Rule-Based Expert System for Resource Planning in Liquid Transportation

Abstract

The dispatch planning process plays a central role in liquid transportation, where the accurate selection of trailers, ISO tanks, vehicles, and drivers determines the effectiveness, safety, and cost structure of operations. Each resource has its own technical, regulatory, and operational characteristics, and these characteristics must align with product specifications, transportation routes, loading and delivery conditions, and the current state of the fleet. The breadth of these parameters makes resource selection a highly complex task for planners, especially in environments where rapid decision-making is needed to address changing demands. This study presents a rule-based expert system designed to capture the decision-making logic of experienced professionals and apply it consistently during dispatch planning. The system incorporates 28 decision rules formulated from the collective knowledge of experts working in liquid logistics operations, including planners, industrial engineers, and senior managers. These rules enable the system to evaluate multiple resource combinations and recommend the most suitable allocation for each order. The expert system was evaluated using real operational data obtained from a leading logistics company in Turkey. Comparative results indicate that the system provides more cost-effective, efficient, and balanced dispatch plans than manual planning conducted by an experienced human planner. The system not only improves resource utilization but also reduces planning errors and variations arising from human judgment. Overall, the findings demonstrate that a rule-based expert system can serve as a reliable and scalable decision-support tool for complex dispatch planning problems in liquid transportation, offering consistent performance across different operational scenarios.

1. Introduction

Logistics holds a highly critical position in the economic structure of a country [1]. The system that enables products to move from the point of production to the point of consumption is called logistics. This system is constantly evolving in various aspects, such as cost management, resource utilization, and time management [2]. Logistics is not limited to transporting and storing products; it also requires effective planning and management [3,4]. In this context, logistics plays a significant role in many sectors, including liquid transportation [5,6].

Liquid transportation is a specialized branch of logistics that deals with the movement of fluid products between locations. Unlike dry cargo transportation, liquid logistics involves additional domain-specific constraints such as product compatibility, tank cleaning requirements, hazardous material regulations (e.g., ADR compliance), and operational restrictions related to equipment suitability. These constraints significantly limit feasible resource combinations and impose strict elimination criteria during dispatch planning. Decision-making in liquid transportation relies heavily on expert judgment developed through experience, where operational uncertainty is typically managed through categorical rules related to product compatibility, tank suitability, ADR compliance, and operational constraints rather than through explicit modeling of continuous physical phenomena such as sloshing effects or viscosity variations. In this study, liquid transportation refers to the transport of products in liquid form, including food-grade liquids and liquid cleaning or hygiene-related products, as well as hazardous liquids. Petroleum-based products and liquefied gases are explicitly excluded from the scope of the study. This method relies on various parameters to ensure safety, cost-effectiveness, and timely delivery, making it an operationally complex structure [7]. Effective planning in liquid transportation is essential for optimizing costs and improving efficiency in logistics operations [8]. Liquid dispatch planning involves key components that affect the cost of operations, productivity, quality, and flexibility within production and logistics facilities.

Efficient dispatch planning also has important implications for sustainable logistics operations. Inefficient resource allocation in liquid transportation may lead to unnecessary vehicle movements, increased fuel consumption, and additional operational costs. By ensuring that appropriate combinations of trailers, ISO tanks, vehicles, and drivers are selected for each order, dispatch planning systems can improve resource utilization and reduce operational inefficiencies. Therefore, improving dispatch planning processes contributes not only to operational efficiency but also to more sustainable transportation practices.

Generally, four types of resources are utilized to transport orders in liquid transportation: trailers, ISO tanks, vehicles, and drivers. The crucial role of these resources in liquid transportation is evident; without them, it is impossible to move products between transportation points. Each resource has unique characteristics that must be considered during selection (e.g., fuel efficiency, dangerous goods transport suitability status, model, etc.), and these characteristics, combined with their ability to be used efficiently under different conditions, make the selection problem extremely broad and complex. The optimal utilization of resources in transporting the ordered products is essential. If appropriate resources are not selected, the products cannot be transported safely, efficiency decreases, delays occur in transportation, and unnecessary costs arise. The smooth execution of the process in liquid transportation depends on the optimal selection of resources. The optimal selection of resources significantly contributes to the efficiency, safety, and cost-effectiveness of the process.

The resource planning process requires experienced experts. Experts develop a plan by considering all possible variables during planning. However, considering these variables is highly complex due to the risk of error, time consumption, and the difficulty of accounting for each variable. In the resource planning process, factors such as the type and quantity of the product requested by the customer, the loading location, the delivery location, and the current state of available resources further complicate the process. This complexity poses a significant challenge in ensuring that the correct resources are used at the right time to respond quickly and flexibly to incoming orders while also enhancing the safety and efficiency of the transportation.

1.1. Literature Review on Transportation

Liquid transportation planning requires the coordinated management of multiple interdependent resources, such as vehicles, containers, and human operators, under strict operational and regulatory constraints. Accordingly, prior studies have investigated various decision-support and optimization-based approaches to address different aspects of this complexity. The existing literature covers a wide range of topics, including tanker vehicle selection under hazardous conditions [9], integrated production–distribution and transportation planning in industrial gas supply chains [10], truck–cargo matching and trailer–tractor separation strategies [11], and multi-objective optimization for integrated harvesting and transportation operations [12], and exact optimization models for long-distance full-truckload liquid transportation networks [13]. In addition, intermodal rail–road transportation models for perishable and liquid-like products have been investigated with an emphasis on environmental and economic sustainability [14], dynamic full-truckload vehicle assignment and routing problems considering heterogeneous fleets, time windows, and multiple trailer and freight types [15], safety-oriented optimization of chemical supply chains [16], and energy-efficient monitoring of liquid goods logistics using agent-based approaches [17]. Research on transportation systems also includes studies on vehicle chassis and suspension technologies that influence the safety and operational performance of heavy-duty vehicles used in logistics operations [18,19].

Beyond these liquid transportation-specific studies, transportation decision-making has also been examined from broader planning and policy perspectives. Several studies focus on transportation mode selection and sustainability-oriented decision frameworks. For example, sustainable transportation evaluation using fuzzy decision-making methods [20], stakeholder-based decision-making approaches for sustainable urban logistics planning [21], freight transportation planning and network optimization models [22], multicriteria decision models for road freight transport service selection [23], and routing optimization approaches for urban freight transportation systems [24] have been proposed to support strategic transportation decisions.

For instance, Zhang et al. [25] examined the mode choice problem in maritime cold-chain logistics by comparing containerized transport and reefer bulk shipping alternatives considering both environmental and economic performance. Unlike these studies that focus on selecting among alternative transport modes, the present study addresses operational dispatch planning within a single transportation mode by determining feasible combinations of transportation resources. Cargo composition optimization problems have also been investigated in transportation planning literature. Recent studies examine how different cargo types can be optimally combined within transportation units to improve capacity utilization and operational efficiency under compatibility and capacity constraints [26]. These studies typically formulate the problem as a mathematical optimization model that determines the optimal mixture of cargos within a transport unit.

In contrast, the present study focuses on operational dispatch planning by selecting feasible combinations of heterogeneous transportation resources, including trailers, ISO tanks, vehicles, and drivers, under regulatory and operational constraints. While previous studies highlight the importance of integrated resource planning and advanced modeling techniques in transportation systems, comprehensive decision-support frameworks that address all relevant operational resources in a unified and operationally explainable manner remain relatively limited in the existing literature.

1.2. Literature Review on Expert Systems

Expert Systems (ESs), since their inception, have found a wide range of applications in various fields, such as pattern recognition, automation, computer vision, virtual reality, diagnostics, image processing, nonlinear control, robotics, automated reasoning, data mining, process planning, and manufacturing [27]. ESs have contributed to the literature in logistics-related areas through various scientific and technical studies [28,29]. In addition, numerous studies have applied expert systems to support logistics operations such as inventory management [30,31,32,33], vehicle route planning [34,35,36], warehousing, production/distribution systems [37,38,39,40,41], and logistics [42,43,44,45,46,47,48]. Furthermore, expert systems have been combined with metaheuristic algorithms to address domain-specific combinatorial optimization problems; for instance, Hakli et al. [49] integrated a genetic algorithm with an expert system for land redistribution planning, demonstrating that the hybrid approach produced applicable redistribution plans in significantly shorter time compared to manual technician-based methods.

1.3. Literature Review on Expert Systems in Resource Planning Problems

The use of ESs in resource planning processes is of great importance in enhancing decision-making efficiency and optimizing operational productivity. In this context, various studies have examined how ESs are utilized in resource planning and the benefits they provide. Prior studies have demonstrated the use of ESs in equipment and resource selection across various domains, including construction, mining, manufacturing, and material handling. These applications cover earthmoving equipment selection [50], equipment optimization in open-pit mining under uncertainty [51], machine selection in manufacturing systems [52], and material selection for sustainable product and process design [53,54,55]. In addition, ES-based approaches have been employed to support material handling equipment selection, aiming to enhance information utilization and resource diversity [56]. Overall, the literature highlights the effectiveness of expert systems in supporting resource planning and selection decisions in complex operational environments. However, despite these contributions, the application of expert systems to operational dispatch planning problems in liquid transportation remains largely unexplored. In particular, studies that formalize expert dispatch planning knowledge into structured rule-based decision-support frameworks for selecting heterogeneous transportation resources under strict operational and regulatory constraints are still limited.

1.4. Motivation of the Study

1.4.1. Research Gap

The recent literature on transportation and logistics planning can be broadly classified into three main categories: (i) optimization-based models focusing on cost, time, and environmental objectives under deterministic or uncertain parameters, (ii) heuristic and meta-heuristic approaches designed to address large-scale and computationally intensive routing and allocation problems, and (iii) uncertainty-aware mathematical frameworks employing fuzzy, intuitionistic fuzzy, or neutrosophic theories to model imprecise data. These streams provide valuable contributions at strategic and tactical decision levels but predominantly emphasize mathematical formulation and numerical solution quality. Studies addressing operational dispatch planning problems that require rapid, explainable, and rule-driven decisions based on expert knowledge remain relatively limited, particularly in liquid transportation settings characterized by strict safety and compatibility constraints.

Most existing studies on liquid transportation planning rely on optimization, heuristic, and meta-heuristic approaches to address large-scale routing and allocation problems. These approaches have demonstrated strong performance in generating near-optimal solutions, especially in complex and resource-intensive transportation environments. Their primary focus, however, remains on computational efficiency and solution quality rather than on transparent and explainable decision-making under strict operational constraints.

In practical liquid transportation operations, resource planning requires the simultaneous consideration of heterogeneous resources, including trailers, ISO tanks, vehicles, and drivers, along with diverse customer demands, product characteristics, delivery locations, and real-time resource availability. This multidimensional structure, combined with strict safety, compatibility, and regulatory constraints, renders the problem computationally intensive and exhibits NP-hard characteristics due to its combinatorial structure, heterogeneous resource types, and multiple operational and compatibility constraints. To illustrate the scale of this combinatorial complexity, consider a representative operational setting with 214 available trailers, 187 ISO tanks, 184 vehicles, and 164 drivers. Even without considering feasibility constraints, the Cartesian product of these resource sets yields more than 1.2 billion possible resource combinations. Exploring such a vast solution space to identify optimal or near-optimal plans within limited decision times is therefore impractical using exhaustive optimization approaches, particularly in dynamic operational environments. From a theoretical perspective, this type of constrained combinatorial selection problem is structurally related to well-known NP-hard formulations in the location and resource allocation literature. For example, the p-median problem selects a fixed number of facilities from a candidate set and assigns demand points to them under combinatorial constraints, and has been formally shown to be NP-hard [57,58]. Although the objective and decision variables differ, both problems share the fundamental characteristic of selecting feasible combinations from large discrete sets under constraints, leading to exponential growth of the solution space.

Simple heuristic decision rules are commonly used in transportation planning to enable fast decisions under time pressure. Such heuristics typically rely on limited criteria, such as proximity, urgency, or workload measures, and aim to simplify the decision process by prioritizing a small subset of easily measurable attributes. While these approaches can be effective in routing or customer sequencing problems, they are not well-suited to liquid transportation dispatch planning, where feasibility is governed by strict safety, compatibility, and regulatory constraints. In the studied problem, a decision cannot be made solely based on distance or urgency, as infeasible resource combinations (e.g., incompatible product–tank pairs or non-compliant ADR resources) must be eliminated before any performance-related consideration.

Moreover, even when heuristic rules select technically feasible resources, repeated application of such simplified selection criteria may lead to systematic overuse of certain trailers, tanks, vehicles, or drivers, while other available resources remain underutilized. Over time, this results in imbalanced resource utilization, increased wear, and reduced operational fairness, which are critical concerns in liquid transportation operations. Therefore, purely heuristic selection rules are insufficient to guarantee safe, balanced, and sustainable dispatch decisions, motivating the need for a rule-based expert system that explicitly encodes domain knowledge, feasibility constraints, and workload-balancing considerations.

In such environments, dispatch decisions are largely shaped by tacit knowledge accumulated by experienced planners, incorporating practical judgments related to resource compatibility, operational risks, and workload balancing. While this experience-based reasoning plays a central role in real-world dispatch planning, it is difficult to capture using conventional optimization models. Despite its importance, the formal integration of expert knowledge into decision-support systems for liquid transportation dispatch planning remains limited in the existing literature. Studies on expert systems in logistics tend to focus on general transportation or material handling problems, with limited attention given to the specific constraints of liquid transportation. As a result, structured decision-support frameworks that formalize expert reasoning and provide consistent, rule-based planning tailored to liquid transportation operations remain largely unexplored.

Based on this research gap, this study investigates whether expert knowledge used in liquid transportation dispatch planning can be systematically formalized into a rule-based decision-support framework capable of generating feasible, consistent, and operationally reliable resource allocation decisions under complex operational constraints.

Although the proposed expert system provides a structured and explainable decision-support framework for dispatch planning, the present study does not include a direct comparison with mathematically optimal solutions obtained from optimization-based models such as mixed-integer programming formulations. Instead, the evaluation focuses on comparisons with manual dispatch planning practices used in real operational environments.

1.4.2. Main Contributions of the Study

This study aims to achieve low-cost and reliable resource allocation in liquid transportation by utilizing a rule-based expert system. The proposed expert system supports dispatch planning by formalizing planner knowledge into predefined rules and facts, enabling the generation of suitable alternative plans for incoming orders. By mimicking the decision logic applied by experienced dispatch planners, the system provides consistent and explainable decision support for operational planning.

The proposed rule-based expert system contributes to regulatory compliance by enforcing mandatory safety, eligibility, and compatibility rules during dispatch planning. Resources that do not satisfy ADR requirements, certification conditions, or product compatibility constraints are automatically excluded, reducing the likelihood of regulatory violations caused by human oversight. The consistent selection of appropriate and well-maintained equipment improves operational reliability and reduces risks associated with improper resource allocation in hazardous liquid transportation.

This study addresses the identified research gap by introducing an expert system–based decision-support framework specifically designed for liquid transportation. The primary contributions of the study are summarized as follows:

• Introduction of a rule-based expert system for dispatch planning in liquid transportation, providing a structured decision-support mechanism for resource selection and allocation.
• Support for human-centered decision-making in dispatch planning by integrating expert knowledge into a transparent rule-based system that assists planners in generating consistent and explainable operational decisions.
• Evaluation of expert system performance using real-world data, based on a leading logistics company in Turkey, reveals superior cost efficiency and more balanced resource utilization compared to manual planning.
• Development of a scalable decision-support system applicable not only to liquid transportation but also to various logistics sectors.

This paper is organized as follows: Section 2.1 introduces the dispatch planning problem in liquid transportation and presents the dataset. Section 2.2 explains the ES. In Section 3, the solution to the dispatch planning problem using the ES is demonstrated. Section 4 evaluates the performance of the ES in the dispatch planning problem. Finally, Section 5 concludes the paper.

2. Materials and Methods

2.1. Resource Planning in Liquid Transportation

Transportation holds significant economic importance worldwide [1]. A subfield of transportation, liquid transportation, plays a critical role in the safe and efficient transport of fluid products. Turkey also plays a vital role in this context, as its geographical location, providing access to waterways such as the Mediterranean, Black Sea, and Aegean Sea, and serving as a bridge between Asia and Europe, offers a strategic advantage for transportation [59,60].

Resources such as ISO tanks, trailers, and vehicles are utilized in the transportation of fluid products. The relevant data were provided by a leading logistics company, and the analysis of the process, including the challenges and constraints in planning, was carried out with the assistance of this company.

Driver: The individual responsible for safely transporting liquid products. Drivers typically work with resources such as ISO tanks, trailers, and vehicles, and they are accountable for the secure transportation of fluid materials. These drivers are employed by logistics companies and undergo a specific training process. While performing tasks such as loading products into tanks, transporting them, and unloading, drivers must adhere to established protocols.

ISO Tank: A transportation tank manufactured according to international standards. These tanks are used to safely transport fluid products. Produced in compliance with standards set by the ISO (International Organization for Standardization), these tanks are designed to carry liquid or gaseous products under high pressure.

Trailer: A vehicle that carries the tank and is towed by a vehicle. In liquid transportation, trailers are generally available in two different types. Standard trailers are those where a tank is mounted onto a trailer and can be towed by a vehicle using special connection points. Tanker trailers, on the other hand, are specially designed for the transportation of fluid products and are an integrated design combining the tank and trailer. Some trailers are manufactured to comply with ADR (Accord Dangereux Routier) regulations, which define the international standards for transporting hazardous materials by road in Europe.

Vehicle: A motor vehicle that tows trailers or other transportation units. A vehicle typically has a high-powered engine and a section to connect auxiliary transportation resources. In liquid transportation, vehicles are used to tow resources like trailers and tankers, ensuring the safe transportation of liquid products. Vehicles are considered a crucial component in the liquid transportation industry to enhance the efficiency of transport processes. Certain vehicles are ADR-certified, meaning they meet the necessary safety and regulatory requirements for transporting hazardous materials under international road transport regulations.

The attribute information of the resources used in liquid transportation is provided in

Table 1. Detailed Overview of Resource Information.

Driver	Iso-Tank	Trailer	Vehicle
Department	Department	Department	Department
ID	ID	ID	ID
Name Surname	Company Name	License Plate	License Plate
State	Vehicle Type	Trailer Type	Vehicle Type
Date of Birth	Tank Number	Transportable Weight	Vehicle Body Type
License Class	Volume	Ownership Status	ADR Status
Driver’s License Expiry Date	Tank Volume	Product Group	Brand
Src 5 License	Tank Tare Weight	ADR Status	Model
Type	Product Group		Fuel Efficiency
			Ownership Status
			Product Group

.

In liquid transportation operations, dispatch planning requires assigning available resources such as vehicles, ISO tanks, trailers, and drivers to incoming transportation orders. A transportation assignment typically consists of several sequential operational steps including vehicle departure from the fleet base, loading at the terminal, delivery to the destination, and return for the next assignment.

In this study, the operational movement associated with a completed transport assignment is referred to as a “position”. A position represents a completed transport task performed by a resource during the planning horizon and is used as an indicator of resource utilization. Figure 1 illustrates the position-based movements occurring during a typical liquid transportation assignment.

As illustrated in Figure 1, a typical transportation assignment begins when a customer order is received. The assigned vehicle departs from the fleet base (garage) and travels to the loading terminal (Position 1). After loading the cargo, the vehicle proceeds to the delivery location (Position 2). Once the delivery operation is completed, the vehicle returns to the fleet base or becomes available for the next dispatch assignment (Position 3).

The position count refers to the number of completed transport assignments performed by a resource within a specified evaluation period. This indicator is used in the proposed expert system to balance resource utilization and to prevent excessive use of the same equipment during dispatch planning.

2.2. Expert System

ESs are defined as computer programs that have the ability to make decisions by mimicking domain experts to solve problems [27,61]. In other words, ESs reach solutions by performing steps such as perception, interpretation, reasoning, learning, communication, and decision-making on problems [27]. The design of an ES is carried out by a team consisting of three main groups: domain experts, knowledge engineers, and software developers [62]. Figure 2 illustrates the main components and functioning of an expert system.

ESs generally consist of three main components: a database, a knowledge base, an inference engine, and an interface.

1.

Database: This is where the data related to the field of study is stored. The stored data may include case studies where the ES is applied, actual situations, or variables specific to the field. The database helps the ES learn and make decisions by retaining past experiences in the field of study. The data stored in the database is kept in a specific format and is then used by the knowledge base and inference engine.

2.

Knowledge Base: The knowledge base plays a crucial role in the successful operation of ESs. It represents a storage and processing section that contains the information, rules, and relationships typically provided by domain experts [63]. The primary factor contributing to the success of ESs in various fields is their reliance on a broad spectrum of knowledge [27]. Solving complex problems requires a significant amount of information, and the ES can achieve successful results by utilizing this knowledge effectively. Consequently, researchers and organizations use ESs to develop and implement various systems across different fields [61]. Domain experts are interviewed to acquire knowledge. This process is quite laborious and time-consuming. However, when knowledge is successfully obtained from experts and applied to the ES, the system can assume a consultative role, guiding and providing information to users by leveraging expert knowledge [50]. In this way, the knowledge base supports the ES’s ability to act as a consultant, enabling researchers and organizations to solve complex problems and develop systems in various fields [61]. The knowledge base can be examined under various classification systems, including Rule-Based Systems (RBSs), Frame-Based Systems (FBSs), Object-Oriented Systems (OOSs), and Case-Based Reasoning (CBR) systems [27]. In this study, the widely used RBS was employed.

3.

Inference Engine: Also known as the interpreter, the inference engine has the ability to make logical inferences using the information in the knowledge base, making it a critical component that constitutes the “brain” of ESs. The inference engine applies rules to inputs and the information in the knowledge base to derive conclusions. In this way, it provides information to the user, evaluates various situations, and offers solutions to problems.

4.

Interface: The interface facilitates communication between the user and the ES. It allows users to input data, ask questions, and receive responses. The interface processes user inputs, transmits them to the inference engine, and presents the obtained results in a comprehensible manner. It can be implemented in various formats, such as graphical interfaces, text-based systems, or voice-command systems.

2.3. Mathematical Formulation of the Dispatch Planning Problem

The dispatch planning problem aims to determine a feasible combination of heterogeneous resources under operational, safety, and regulatory constraints. The objective of the proposed approach is not to compute a mathematically optimal solution, but to generate feasible, explainable, and operationally consistent resource allocations within limited decision times. Let an incoming transportation order be denoted by $o$.

2.3.1. Sets

To formally describe the resource structure of the dispatch planning problem, the main resource sets used in the model are summarized in

Table 2. Sets used in the mathematical formulation of the dispatch planning problem.

Symbol	Description
$T$	Set of available trailers
$I$	Set of available ISO tanks
$V$	Set of available vehicles
$D$	Set of available drivers
$O$	Set of transportation orders
$T^{\prime }$	Reduced feasible trailer set after rule filtering
${\mathrm{I}}^{″}$	Reduced feasible ISO tank set after rule filtering
$V^{\prime }$	Reduced feasible vehicle set after rule filtering
$D^{\prime }$	Reduced feasible driver set after rule filtering
$\mathsf{\Omega }$	Set of feasible resource combinations generated from $T^{\prime }\times I^{\prime }\times V^{\prime }\times D^{\prime }$

.

Each incoming transportation order o ∈ O is characterized by attributes including product group PG(o), ADR requirement ADR(o) ∈ {0,1}, transportation distance type Dist(o) ∈ {Short, Long}, and loading location L(o).

2.3.2. Decision Variable

To represent the assignment of resources to a transportation order, the following binary decision variable is defined:

$$x_{t,i,v,d}=\left\{\begin{array}{cc}1& \mathrm{if} \mathrm{trailer} t, \tan \mathrm{k} i, \mathrm{vehicle} v, \mathrm{and} \mathrm{driver} d \mathrm{are} \mathrm{assigned} \mathrm{to} \mathrm{order} o\\ 0& \mathrm{otherwise}\end{array}\right.$$

The dispatch planning problem addressed in this study focuses on the selection and allocation of compatible transportation resources for a given transportation order. Shipment quantities are predefined by customer orders and therefore are not decision variables of the model.

For each incoming order, the decision-making process determines a feasible combination of resources consisting of a trailer, an ISO tank, a vehicle, and a driver. Accordingly, the solution to the dispatch planning problem can be represented as a tuple $x_{t,i,v,d}$, where t ∈ T denotes the selected trailer, i ∈ I denotes the selected ISO tank, v ∈ V denotes the selected vehicle, and d ∈ D denotes the selected driver.

The dispatch planning problem therefore seeks to select feasible combinations of trailers, ISO tanks, vehicles, and drivers that satisfy operational constraints while minimizing the evaluation score.

2.3.3. Operational Constraints

In the proposed expert system, each rule is modeled as an eligibility constraint that filters infeasible resources before combinatorial evaluation. These constraints are applied sequentially to reduce the decision search space while preserving operational feasibility.

For orders requiring ADR-compliant transportation, only resources satisfying ADR requirements are considered:

$$ADR\left(r\right)\ge ADR\left(o\right),\forall r\in \{T\cup I\cup V\cup D\}$$

(1)

ISO tanks must be compatible with the product group of the order:

$$Compat(i,PG(o\left)\right)=1,\forall i\in I$$

(2)

Only resources that are currently available are eligible for selection:

$$Status\left(r\right)=Available,\forall r\in \left\{T,I,V,D\right\}$$

(3)

To ensure fair and balanced resource utilization, average-based workload constraints are applied. For vehicles, let $\mathrm{k}\mathrm{m}\left(v\right)$ denote the distance traveled by vehicle $v$ in the last month, and let the average distance be defined as:

$$\stackrel{ˍ}{km}={\displaystyle \frac{1}{\mid V\mid }}\sum _{v\in V}km\left(v\right)$$

(4)

Eligible vehicles must satisfy:

$$km\left(v\right)\le \stackrel{ˍ}{km}$$

(5)

Similarly, ISO tanks are evaluated based on their recent position counts to avoid excessive reuse.

Vehicle model year constraints are applied based on the transportation distance:

$$ModelYear\left(v\right)\ge \theta \left(Dist\right(o\left)\right)$$

(6)

where $\theta$ denotes a distance-dependent threshold determined by expert knowledge. After applying all eligibility constraints, reduced feasible resource sets are obtained:

$$T^{\prime }\subseteq T,I^{\prime }\subseteq I,V^{\prime }\subseteq V,D^{\prime }\subseteq D$$

(7)

2.3.4. Combinatorial Solution Space

Feasible resource combinations are then generated from the Cartesian product:

$$\mathsf{\Omega }=T^{\prime }\times I^{\prime }\times V^{\prime }\times D^{\prime }$$

(8)

The size of the feasible solution space grows combinatorially with the number of available resources. Even after rule-based filtering, the dispatch planning problem exhibits NP-hard characteristics due to the exponential growth of possible resource combinations.

2.3.5. Evaluation Function

Each feasible solution $x\in \mathsf{\Omega }$ is evaluated using a composite penalty function:

$$Score\left(x\right)={\displaystyle \sum }_{j=1}^p{w}_j{f}_j\left(x\right)$$

(9)

where $f_j\left(x\right)$ represents penalty components such as mileage deviation, fuel efficiency, and workload balance, and $w_j$ denotes expert-defined weights. Final recommendations are obtained by ranking feasible solutions according to their penalty scores.

The rule-based decision procedure used in the proposed expert system follows a sequential filtering and evaluation mechanism. First, operational constraints are applied to eliminate infeasible resources from the candidate sets. The remaining feasible resources are then combined to generate candidate transportation plans. Each candidate plan is evaluated using the penalty-based scoring function defined in Equation (9). Finally, feasible combinations are ranked according to their penalty scores and the most suitable resource allocation is recommended to the dispatch planner.

The detailed pseudocode of the rule-based decision procedure is provided in Algorithm A1 in Appendix A. A conceptual flow diagram illustrating the main steps of the rule-based dispatch planning procedure is presented in Figure 3.

3. Implementation of Expert System for Liquid Transportation

The system development process involved extensive consultations with liquid transportation experts from a well-established logistics company. The primary objective of ESs is to encode expert knowledge in a form that can be easily understood by general users, to establish an inference mechanism, and to represent this knowledge as rules. The logic of the system consists of a series of rules connected by “if-then” reasoning.

3.1. Knowledge Acquisition Process

Tacit knowledge from experienced liquid logistics dispatchers was elicited through structured interviews, supported by the analysis of historical dispatch decisions and iterative validation sessions. The underlying expert reasoning was examined by breaking it down into concrete decision elements, such as eligibility conditions, exclusion criteria, and prioritization rules. These elements were subsequently encoded as if–then production rules, allowing the expert system to reflect expert-level reasoning during resource allocation. The knowledge acquisition process is one of the most crucial components of an ES [64]. In the ES, the knowledge base contains information about resources, movements, laws, materials, and operations. Such information is referred to as attributes, and each has various values. An ES attempts to match attribute values with resource characteristics through its rules to identify the appropriate resource for the incoming order’s requirements.

In this study, knowledge was gathered from various experts and industrial engineers in the field of planning within liquid transportation. A major Turkish logistics firm employs highly knowledgeable and experienced personnel in the transportation industry. The experiences of these knowledgeable staff members played a critical role in determining which resource should be used for each operation within the ES.

A team of seven planners, recognized as experts in liquid dispatch planning, was involved in the design of the ES, bringing extensive experience from a prominent logistics provider. These planners possess deep knowledge and expertise in liquid dispatch planning. Knowledge engineers, senior managers, and technical staff are of significant importance in the design of the ES. These individuals come together to assess all details of the planning processes and develop solutions, ensuring the successful design of the ES by leveraging the expertise of each team member.

Addressing any problem is not limited to resolving the surface issue; it also involves understanding the constraints, demands, and conflicts surrounding the problem, which is of critical importance. Effectively identifying constraints, demands, and conflicts forms the foundation of a successful problem-solving process. Proper implementation of these steps is crucial for completing the work successfully and overcoming various challenges. This section focuses on identifying constraints, demands, and conflicts within the context of the resource selection problem encountered in liquid transportation. Correctly defining these elements is key to a successful solution process.

Balancing constraints, demands, and conflicts is essential for effective problem resolution. Adapting constraints and demands to each other, managing conflicts, and ensuring balance play a crucial role in the problem-solving process. Expert planners in the field conducted a detailed analysis of the current system, meticulously identifying and recording the constraints that affect the dispatch planning process.

The list of identified constraints and demands is shown in

Table 3. List of Constraints, Demands, and Conflicts.

Group	Restriction Demand Number	Content of Restriction/Demand	Areas of Impact	Conflict Contradiction
Law	1	The transported product must comply with domestic transportation laws	ISO Tank	-
ISO Tank	2	Appropriate ISO Tanks must be selected according to the product group planned for transportation.	Trailer, Vehicle, ISO Tank, and Driver	3, 4, 5, 20
	3	The balance of usage count for the selected ISO tank over the last three months	ISO Tank	2, 16, 17, 20
	4	Ensure at least 9 average positions per month	ISO Tank	2, 16, 17
	5	A maximum of 40% of total positions should be within city positions	ISO Tank	2, 16, 17
Trailer	6	Balanced usage count of all trailers in the last three months	Trailer	16, 17, 20
	7	Ensure at least 9 average positions per month	Trailer	16, 17
	8	A maximum of 40% of total positions should be within city positions	Trailer	16, 17
Vehicle	9	Balance of kilometers traveled by vehicles in the last three months instead of usage count	Vehicle	16, 17, 18
	10	Ensure at least 12,000 km per month	Vehicle	16, 17, 18
Driver	11	Balance the number of hours worked among drivers in the last month	Driver	16, 17
	12	Balance the kilometers traveled among drivers in the last month	Driver	16, 17
	13	Balance the ratio of short and long-distance trips among drivers in the last month.	Driver	16, 17
	14	Drivers should work in the liquid department	Driver
	15	Drivers should have an SRC certificate for transporting flammable materials.	Driver
Fuel Savings	16	Select resources that are close to the order location on the dates of planning.	Trailer, Vehicle, ISO Tank, and Driver	3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13
	17	Assign returning resources to another order if possible	Trailer, Vehicle, ISO Tank, and Driver	3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13
	18	Use low fuel-consuming vehicles for long distances and high fuel-consuming vehicles for short distances.	Vehicle	9, 10
General	19	Select available resources	Trailer, Vehicle, ISO Tank, and Driver	-
	20	Take the detachment costs of connected trailers and tanks into account	Trailer and ISO Tank	2, 3, 6
	21	Consider self-tank trailers during trailer selection	Trailer and ISO Tank	-

. Based on feedback from the planners, the demands and constraints have been categorized under seven main headings: Law, ISO Tank, Trailer, Vehicle, Driver, Fuel Efficiency, and General. Instances where these demands and constraints may conflict with each other are clearly specified in the “Conflict/Contradiction” column. A conflict arises when fulfilling one requirement makes it difficult or impossible to meet another. The identified constraints and demands reflect the practical knowledge used by experienced dispatch planners in daily operations. Each constraint represents a decision rule that planners routinely apply in practice, but that is rarely written down in a structured manner. By expressing this experience as explicit constraints, the expert system supports more consistent and transparent decision-making.

For example:

• If the planner selects an ISO tank that has been frequently used, the usage balance of other tanks will be disrupted. However, selecting less frequently used tanks may lead to transportation safety and compatibility issues. (There is a conflict between Items 2 and 3, 4, 5, 20.)
• When the closest vehicle to the order location is selected, some vehicles will be overutilized while others will remain underutilized, leading to an imbalance in mileage distribution. (There is a conflict between Items 9, 10, and Items 16, 17, 18.)
• A returning driver should ideally be assigned to another order to avoid an empty return trip. However, this may result in excessive working hours for certain drivers. (There is a conflict between Items 11, 12, 13, and Item 17.)
• Reallocating connected tanks may help maintain usage balance. However, this detachment process incurs additional costs. (There is a conflict between Item 20 and Items 2, 3, and 6.)

The “Areas of Impact” column records which elements (such as trailers, vehicles, or drivers) are influenced by each constraint or demand.

3.2. Data Preprocessing

Since the obtained data cannot be directly used in the ES, a preprocessing step was conducted on the data; for example, information such as the total distance a vehicle has traveled in the last month is not directly available as an attribute in the raw data. Therefore, it is necessary to perform calculations using the relevant information. Data tables and the relationships between them are used for this calculation.

The tables and their relationships are shown in Figure 4. As illustrated in the figure, the driver, trailer, vehicle, ISO tank, and order tables are linked to the dispatch table. The product table is also related to the order table. The dispatch table represents the process from loading to delivery of an order. Information such as which equipment is used, from where to where when, and in what order is being transported can be understood from this table.

The total distance traveled, hours, and number of positions for a trailer, vehicle, ISO tank, or driver over the past month are calculated through joint operations on the dispatch table. Similarly, information about their most recent location and availability can also be obtained from these joint operations. By using the obtained facility information, it is expected that costs will be reduced by assigning the closest available equipment to new incoming orders.

The database schema used in the expert system is illustrated using an Entity–Relationship (ER) representation, as shown in Figure 4. The relationships between entities are explicitly indicated using cardinalities such as 1:1 and 1:N to clarify how operational resources are associated within dispatch planning operations.

3.3. Development of the Expert System

The development of the ES involves converting all the experiences and insights of experts in the field of liquid transportation planning into “if-then” statements.

3.3.1. Database

Facts are data sets that contain essential information in liquid transportation. In liquid dispatch planning, these important details for logistics companies include information such as the availability of equipment (available, in transit, etc.), whether the driver has the necessary certifications, and whether the tank is ADR compliant. These details are crucial for ensuring that the dispatch is carried out safely and efficiently.

These facts are used in the ES within the “if-then” rule structure, allowing for the creation of predefined logical rules. For example, if the product to be transported is flammable, rule-based logic can be developed to prevent drivers without a flammable goods transportation certificate from being assigned to the task. In this way, the ES can effectively utilize real-world data to produce the most suitable, safe, and cost-effective solutions in liquid dispatch planning. The main facts used in the proposed expert system are summarized in

Table 4. Facts.

Fact Name	Description
Short Distance	100 km or less
Suitable ISO Tank	An ISO Tank that is suitable for transporting the incoming order
Suitable Trailer	Trailers that are suitable for transporting the incoming order
Suitable Driver	The driver who is suitable for transporting the incoming order
Suitable Vehicle	Vehicles that are suitable for transporting the incoming order
Planning	The Cartesian product of Suitable ISO Tank, Suitable Trailer, Suitable Vehicle, and Suitable Driver variables
Order Loading Location	The location where the product ordered by the customer will be loaded onto the resource
Order Delivery Location	The location where the product ordered by the customer will be delivered
Fuel Efficiency	Information on how much fuel the vehicle consumes
Vehicle Model	The year of the vehicle model
Available ISO Tank	ISO Tank that is clean and idle in the garage
Available Trailer	Trailer that is clean and idle in the garage
Available Vehicle	Vehicle that is clean and idle in the garage
Available Driver	The driver who is idle and available in the garage
Product Group	Group information of the ordered product
Flammable Product:	Products with an ADR label are considered flammable.
Self-Tank Trailer	Trailers where the tank and trailer are combined and inseparable
Nearest First garage	The garage closest to the loading location
Nearest Second garage	The second closest garage to the loading location
Position Count	The movement of resources from one place to another is referred to as a position.
Intra-City Order	If the loading and delivery locations are in the same city, it is an intra-city order.
Inter-City Order	If the loading and delivery locations are in different cities, it is an inter-city order.
Product Category	Classification of liquid products (e.g., food-grade, hazardous, volatile chemical)
Tank Cleaning Status	Indicates whether the ISO tank has been cleaned and certified for the next product category

.

3.3.2. Preparation of the Rule Base

Experts and system developers have come together to design a set of rules that provide expertise on a specific domain or topic. These rules are used to allow the system to evaluate certain situations and make decisions. Each rule defines a situation and specifies a decision to be made under that condition. These rules were created to reflect how dispatch planners actually make decisions, rather than to optimize purely mathematical objectives. In the rule base, product groups are used to distinguish between different types of liquids, such as food-grade products and hazardous or volatile chemicals. The compatibility between products and ISO tanks is assessed based on product group information, suitability conditions, and tank cleanliness status. In food-grade liquid transportation, hygiene and contamination risks are primarily controlled through certified tank cleaning procedures performed before dispatch planning. ISO tanks are required to undergo standardized cleaning and inspection processes to remove residues from previous cargos. As a result, cross-contamination risks are operationally mitigated through tank preparation procedures prior to resource assignment. Within the proposed expert system, product group compatibility and tank suitability conditions ensure that only appropriate tanks are considered during the dispatch decision process. By applying these product group–based rules, the system naturally avoids cross-contamination while keeping the existing decision structure unchanged.

For example, in the context of liquid dispatch planning, the rule base may include rules such as the following:

• If the liquid product to be transported is a flammable substance, then select a driver who is certified to transport flammable materials.

This rule base guides the ES in making decisions and solving problems during liquid dispatch planning. The preparation of the rule base allows the system to effectively utilize knowledge specific to its area of expertise. The main production rules used in the expert system are presented in

Table 5. Production Rule Method (If-Then Rules).

Rule Number	Rules
1	If	There is an available and clean ISO tank	Then	Assign the ISO tanks to the SuitableISOtank variable
2	If	There are ISO Tanks among the suitable ISO tank variables that can transport the ordered product group	then	Select from them and update the SuitableISOtank variable.
3	If	There are available and idle trailers	then	Assign the available trailers to the SuitableTrailer variable.
4	If	The ordered product is a flammable product	then	Select trailers that can transport flammable materials from the SuitableTrailer variable and update it.
5	If	There are self-tank trailers in the SuitableTrailer variable	then	remove trailers that cannot transport the ordered product group from the SuitableTrailer variable.
6	If	There are available and idle vehicles	then	Assign the available vehicles to the SuitableVehicles variable.
7	If	The ordered product is a flammable product	then	Select vehicles that can transport flammable materials from the SuitableVehicles variable and update it.
8	If	There are available and idle drivers	then	Assign the available drivers to the SuitableDrivers variable.
9	If	The ordered product is a flammable product	then	Select drivers that can transport flammable materials from the SuitableDrivers variable and update it.
10	If	There are trailers, ISO Tanks, drivers, and vehicles at the nearest first garage.	then	Assign them to the SuitableTrailer, SuitableISOtank, SuitableDrivers, and SuitableVehicles variables, respectively.
11	If	None of the trailers, ISO Tanks, drivers, or vehicles are at the nearest first garage, but are at the nearest second garage	then	Assign those to the respective variables.
12	If	None are at the first or second yard, but are at the third nearest garage	then	Assign them to the respective variables.
13	If	There are ISO Tanks in the SuitableISOtank variable with a lower average position count in the last month	then	Select them and update the SuitableISOtank variable.
14	If	The order is intra-city, and there are ISO Tanks with a city position rate lower than 40%	then	Bring those ISO Tanks and update the SuitableISOtank variable.
15	If	The order is inter-city, and there are ISO Tanks with an inter-city position rate lower than 60%	then	bring those ISO Tanks and update the SuitableISOtank variable.
16	If	There are trailers in the SuitableTrailer variable with a lower average position count in the last month	then	Select them and update the SuitableTrailer variable.
17	If	The order is intra-city, and there are trailers with a city position rate lower than 40%	then	Bring those trailers and update the SuitableTrailer variable.
18	If	The order is inter-city, and there are trailers with an inter-city position rate lower than 60%.	then	Bring those trailers and update the SuitableTrailer variable.
19	If	There are vehicles in the SuitableVehicles variable with a lower average kilometer traveled in the last month	then	Select them and update the SuitableVehicles variable.
20	If	The distance between the loading and delivery location is more than 100 km	then	Select vehicles with a lower fuel efficiency rate and assign them to the SuitableVehicles variable.
21	If	The distance between the loading and delivery location is less than 100 km	then	Then select vehicles with a higher fuel efficiency rate and assign them to the SuitableVehicles variable.
22	If	The distance between the loading and delivery location is more than 100 km	then	Select newer model vehicles and assign them to the SuitableVehicles variable.
23	If	The distance between the loading and delivery location is less than 100 km	then	Select newer model vehicles and assign them to the SuitableVehicles variable.
24	If	The distance between the loading and delivery location is more than 100 km	then	Select drivers with a lower ratio of trips over and under 100 km and update the SuitableDrivers variable.
25	If	The distance between the loading and delivery location is less than 100 km	then	Select drivers with a higher ratio of trips over and under 100 km and update the SuitableDrivers variable.
26	If	There are drivers in the SuitableDrivers variable with a lower average number of kilometers traveled in the last month	then	Select them and update the SuitableDrivers variable.
27	If	The tank and trailer are combined	then	They will not be combined with other tanks or trailers in the planning variable.
28	If	The trailer is self-contained	then	No ISO Tank matching will be made with it in the planning.

.

The rule thresholds and evaluation indicators were derived through structured knowledge acquisition sessions with experienced dispatch planners and validated using historical operational data. Threshold values such as the 40% intra-city/inter-city ratio, the 100 km distance boundary, and the one- and three-month evaluation windows reflect commonly applied operational heuristics used by planners to balance workload, prevent overutilization, and ensure regulatory and safety compliance in daily dispatch decisions.

Evaluation indicators, including position count, traveled distance, fuel efficiency, vehicle model year, and short–long distance ratios, were selected to represent key operational objectives, such as balanced resource utilization, cost efficiency, equipment wear control, and fair workload distribution among drivers. These indicators correspond to measurable performance factors routinely monitored by logistics firms and were confirmed by domain experts as critical for practical decision-making in liquid transportation operations. By encoding these expert-validated thresholds and indicators into explicit rules, the system ensures transparent, explainable, and operationally grounded dispatch planning.

3.3.3. Inference Engine Design

This step refers to a mechanism that enables the derivation of conclusions using specific rules and facts from the rule base. In other words, it involves the process that allows the ES to arrive at a logical conclusion when faced with a particular situation or problem. The inference mechanism typically operates using an “if-then” rule structure. That is, if certain conditions are met, a specific conclusion is reached. In this case, the ES makes logical inferences by applying the rules from the rule base and the input facts using a forward chaining approach, where the system starts from known facts and applies rules iteratively to determine the best outcome.

For example, consider an ES for liquid dispatch planning. The inference mechanism can generate an appropriate dispatch plan based on the characteristics of the liquid product to be transported, the transportation conditions, and the route information. If the incoming order involves a long-distance journey and a flammable product, the inference mechanism selects the most suitable resources that can travel long distances and transport flammable materials. This step represents the ES’s ability to process complex information and allows the system to attain a level of expertise in a specific domain. The inference mechanism is a critical component that enhances the ES’s problem-solving and decision-making capabilities.

In this study, four different resources are used to obtain planning results, each with a specific quantity. The Cartesian product of these resources forms the basis of the planning process. However, two important considerations are taken into account during this process.

• First, in the case where the trailer and ISO tank are combined, the system presents them as the same resources in the planning process without separating them.
• The second case is when the trailer is a self-contained tanker. Such trailers will not be included in the Cartesian product with ISO tanks; instead, the trailer will undergo a Cartesian product with the other drivers and vehicles separately.

After the Cartesian products were performed, a scoring process was applied to each plan to present the top 10 planning options to the user. Equation (10) was used to score the plans. The plans with the lowest scores are considered to have the best planning.

$$\mathit{Score}(x)=w_1{\displaystyle \frac{q_t}{Q}}+w_2{\displaystyle \frac{s_i}{S}}+w_3{\displaystyle \frac{k_v}{K}}+w_4{\displaystyle \frac{k_d}{X}}+w_5{\displaystyle \frac{f_v}{F}}+w_6{\displaystyle \frac{y_v}{Y}}+w_7\mathsf{\phi }(m_s,m_l,R)$$

(10)

$$\mathsf{\phi }\left(m_s,m_l,R\right)=\left\{\begin{array}{c}{\displaystyle \frac{\left(m_s/m_l\right)}{R}}, if Dist\left(o\right)=Short\\ {\displaystyle \frac{\left(m_l/m_s\right)}{R}}, if Dist\left(o\right)=Long\end{array}\right.$$

(11)

$$R={\displaystyle \frac{1}{\left|X\right|}}{\displaystyle \sum }_{d=1}^X{\displaystyle \frac{m_s^{\left(d\right)}}{m_l^{\left(d\right)}}}$$

(12)

In this equation, $x=\left(t,i,v,d\right)$ represents a feasibility plan. $q_t$ denotes the number of positions for the selected trailer, and $Q$ is the total number of trailer positions in the system. $s_i$ and $S$ represent the number of available positions for the selected ISO tank and the total ISO tank positions, respectively. $k_v$ and $K$ denote the total kilometers traveled by the selected vehicle and by all vehicles in the last month. $k_d$ and $X$ represent the total kilometers driven by the selected driver and by all drivers in the same period. $f_v$ and $F$ correspond to the fuel consumption rate of the selected vehicle and the total fuel consumption rate of the fleet. $y_v$ indicates the model year index of the selected vehicle, and $Y$ is the sum of model year indices for all vehicles. The variables $m_s$ and $m_l$ refer to the number of short- and long-distance trips completed by the driver. The function $\mathsf{\phi }\left(m_s,m_l,R\right)$ is formulated as an adaptive heuristic that penalizes unbalanced short- and long-distance assignments based on the distance category of the current order. By normalizing individual driver trip distributions against the system-level ratio $R$, the scoring function discourages persistent imbalance in trip assignments, promotes balanced workload allocation, mitigates driver fatigue, and supports sustainable resource utilization in liquid transportation operations.

This formula is a weighted average designed to assess various factors in transportation planning. It takes into account the frequency of resource usage, the distance covered by both vehicles and drivers, and the fuel efficiency. Additionally, it incorporates the model year of the vehicles, which could affect maintenance costs and efficiency.

Each component of the formula is normalized by dividing the specific value (e.g., a particular trailer’s position count) by the total available values (e.g., total trailer positions). This allows for a fair comparison and aggregation of different resources, ensuring that no single factor disproportionately influences the overall assessment.

The goal is to achieve optimal dispatch planning by ensuring balanced utilization of trailers, tanks, vehicles, and drivers while also considering fuel consumption and vehicle model age.

3.3.4. Interface

Users can view, sort, and evaluate the system-generated planning recommendations for a specific order through the interface. The interface presents order details, truck and trailer options, driver information, and computed scores to the user. By utilizing filtering and sorting options, users can identify the most suitable resource combinations and examine the proposed plans in detail. Additionally, an “Export to Excel” feature is available, enabling users to export selected plans for further analysis. This functionality allows users to conduct a comprehensive evaluation of the planning process and leverage the expert system’s data-driven insights for informed decision-making. The interface of the proposed expert system is illustrated in Figure 5.

Although the proposed system automatically filters infeasible resource combinations according to predefined rules, the final decision authority remains with the dispatch planner. The interface allows planners to review alternative feasible resource combinations and manually adjust assignments when necessary.

In situations where operational exceptions occur, planners can override the automatically generated recommendation and assign a specific resource manually. In addition, the system interface includes a settings panel that allows users to temporarily deactivate selected rules when required. This functionality enables planners to handle exceptional operational cases while maintaining the transparency of the rule-based filtering process.

These features ensure that the expert system operates as a decision-support tool rather than a fully autonomous decision maker, preserving human control over the dispatch planning process.

4. Results and Discussion

The system was developed using the C# programming language in Visual Studio 2022. It was tested on a local development machine with an Intel Core i5-7200U CPU, 20 GB of RAM, and a 64-bit operating system. The system operated efficiently under these conditions, demonstrating its robustness and suitability for real-world applications. The ES was applied to the order provided in

Table 6. Comprehensive Overview of Order Details.

Number	Product Name	Loading Location	Delivery Location	Product Group	ADR Status	Distance Status
1	VORANATE™ M 229 Polymeric MDI	Dilovasi, Kocaeli, Turkey	M.Kemalpaşa, Bursa, Turkey	MDI	Non-ADR	Long
2	Texapon N 70	Cayirova, Kocaeli, Turkey	Ankara, Elmadag, Ankara, Turkey	Standart	Non-ADR	Long
3	VORANATE T-80 TYPE	Dilovasi, Kocaeli, Turkey	Odunpazari, Eskisehir, Turkey	TDI	ADR	Long
4	DINP	Dilovasi, Kocaeli, Turkey	Cayirova, Kocaeli, Turkey	Standart	Non-ADR	Short
5	CIKOLATA	Odunpazari, Eskisehir, Turkey	Selcuklu, Konya, Turkey	Chocolate	Non-ADR	Long
6	POLIPOL 764 (DOKME)	Crkezkoy, Tekirdag, Turkey	Tuzla, Istanbul, Turkey	Rosin	ADR	Long
7	BUTYL ACRYLATE	Dilovasi, Kocaeli, Turkey	Dilovasi, Kocaeli, Turkey	Standart	ADR	Short
8	LABSA(Linear Alkyl Benzene Sulphonic Acid)	Corlu, Tekirdag, Turkey	Gebze, Kocaeli, Turkey	Standart	ADR	Long

. It includes products to be transported from various cities in Turkey to different destinations, along with the transportation details of these products.

Each order includes important details such as the order code, product name, loading and delivery locations, product group, and ADR status. If the loading and delivery locations are within the same city, this is considered intra-city transportation. For transportation between two cities, if the distance exceeds 100 km, it is categorized as long-distance transportation. This information is critical for dispatch planning.

4.1. Sensitivity Analysis

To examine the robustness of the proposed penalty-based evaluation framework, a sensitivity analysis was conducted by varying the weight coefficients associated with the scoring function components. Several scenarios were designed to emphasize different operational priorities, such as fuel efficiency, driver workload balance, and equipment utilization. The resulting ranked plans were then compared across scenarios to assess the stability of the decision outcomes. Since the penalty weights directly influence the evaluation function defined in Equations (9) and (10), the sensitivity analysis also serves to examine the robustness of the decision framework with respect to different weight configurations. The results of the sensitivity analysis are presented in

Table 7. Sensitivity analysis results under different weight configurations.

Scenario	w1	w2	w3	w4	w5	w6	w7	Top-Ranked Solution Changed?
S1 (Baseline)	1	1	1	1	1	1	1	Baseline reference
S2 (Fuel-focused)	1	1	2	1	2	1	1	Top-ranked solution changed (trailer selection only)
S3 (Driver-focused)	1	1	1	2	1	1	2	Same top-ranked solution as S2
S4 (Equipment-focused)	2	2	1	1	1	1	1	Same top-ranked solution as S2

.

The sensitivity analysis indicates that varying the weight coefficients affects only the relative ordering of alternative plans, primarily through changes in trailer selection. The top-ranked solutions consistently rely on the same ISO tank, vehicle, and driver assignments across all scenarios, demonstrating that the proposed decision framework is robust with respect to reasonable weight variations. No structural change in feasibility or resource compatibility is observed.

4.2. Expert System Application and Analysis on a Single Order

As a result of applying the rules from Table 5 in the ES, the analysis of the four resources involved in the transportation of Order-3 is provided in

Table 8. Changes in Resource Numbers After Each Rule.

Rule Number	Affected Resources	Description	Number of Trailers	Number of ISO Tanks	Number of Vehicles	Number of Drivers
	-	Initial values	214	187	184	164
1	ISO Tank	Containers that are full are eliminated	214	177	184	164
2	ISO Tank	Elimination based on product group	214	5	184	164
3	Trailer	Trailers that are full are eliminated	202	5	184	164
4	Trailer	Elimination based on ADR information	176	5	184	164
5	Trailer	Elimination based on product group for self-tank trailers	111	5	184	164
6	Vehicle	Vehicles that are full are eliminated	111	5	177	164
7	Vehicle	Elimination based on ADR information	111	5	92	164
8	Driver	Drivers who are full are eliminated	111	5	92	157
9	Driver	Elimination based on ADR information	111	5	92	66
10, 11, 12	All	Resources close to the order’s loading location are selected	80	5	71	66
13	ISO Tank	Containers with position counts below the average are selected	80	2	71	66
14, 15	ISO Tank	Elimination based on a 40% intra-city position rate for containers	80	2	71	66
16	Trailer	Trailers with position counts below the average are selected	44	2	71	66
17, 18	Trailer	Elimination based on a 40% intra-city position rate for trailers	8	2	71	66
19	Vehicle	Vehicles with kilometers below the average in the last month are selected	8	2	32	66
20, 21	Vehicle	Elimination based on fuel efficiency for long and short distances	8	2	20	66
22, 23	Vehicle	Vehicle model selection based on order distance	8	2	8	66
24, 25	Driver	Drivers are selected based on their long-distance and short-distance ratios	8	2	8	48
26	Driver	Drivers with kilometers below the average in the last month are selected	8	2	8	25

. Order-3, listed in Table 6, refers to the transportation of “VORANATE T-80 TYPE” from Kocaeli, Dilovası to Eskişehir, Odunpazarı. The evaluations related to Table 8 demonstrate how different types of resources were eliminated based on specific rules and how this ultimately led to a reduction in the number of available resources.

Initially, the system contained 214 trailers, 187 ISO tanks, 184 vehicles, and 164 drivers. After the rules were applied sequentially, the number of available resources was significantly reduced.

The initial number of ISO tanks was 187. After Rule-1, which removes containers that are already full, the number decreased to 177, corresponding to a 5.35% reduction. With Rule-2, filtering based on product group reduced the number of ISO tanks to 5, resulting in a 97.18% decrease. This was the highest reduction observed for ISO tanks. Rule-13, which selects containers with position counts below the average, reduced the number further to 2, corresponding to a 60% reduction. Rules 14 and 15, applied using a 40% intra-city transport threshold, did not change the number of ISO tanks.

The initial number of trailers was 214. After Rule-3, which eliminates full trailers, the number decreased to 202, resulting in a 5.61% reduction. Rule-4, based on ADR information, reduced the number to 176, corresponding to a 12.87% decrease. Rule-5, which removes self-tank trailers according to product group compatibility, reduced the number to 111, achieving a 36.93% reduction. Rule-16, which selects trailers with position counts below the average, reduced the number to 44, corresponding to a 60.36% decrease. Rules 17 and 18, applied with a 40% intra-city threshold, reduced the number of trailers to 8, resulting in an 81.82% reduction.

The initial number of vehicles was 184. After Rule-6, which removes vehicles that are already full, the number decreased to 177, corresponding to a 3.80% reduction. Rule-7, based on ADR compatibility, reduced the number of vehicles to 92, resulting in a 48.02% reduction. Rule-19, which selects vehicles with below-average traveled distance in the last month, reduced the number to 32, corresponding to a 65.22% decrease. Rules 20 and 21, applied based on fuel efficiency for long- and short-distance operations, reduced the number to 20. Rules 22 and 23, which perform vehicle model selection according to order distance, further reduced the number of vehicles to 8, resulting in a 60% reduction.

The initial number of drivers was 164. After Rule-8, which removes drivers who are already assigned, the number decreased to 157, corresponding to a 4.27% reduction. Rule-9, based on ADR qualifications, reduced the number of drivers to 66, resulting in a 57.96% decrease. Rules 24 and 25, which evaluate drivers according to long- and short-distance driving ratios, reduced the number to 48, corresponding to a 27.27% reduction. Rule-26, which selects drivers with below-average traveled distance in the last month, reduced the number further to 25, resulting in a 47.92% decrease.

As a result, the total number of resources was substantially reduced. The number of trailers decreased from 214 to 8, ISO tanks from 187 to 2, vehicles from 184 to 8, and drivers from 164 to 25. The most effective rule for ISO tanks was Rule-2, while Rules 17 and 18 were the most effective for trailers. For vehicles, the highest reduction was achieved by Rule-19, and for drivers by Rule-9. These results show that the applied rules efficiently reduced the search space and improved resource selection efficiency.

Considering the Cartesian product formulation defined in Equation (8), the initial theoretical solution space included all possible combinations of the available resources. At the beginning of the process, the system contained 214 trailers, 187 ISO tanks, 184 vehicles, and 164 drivers. This corresponds to approximately 1.2 billion potential resource combinations. After the sequential application of the rule-based filtering mechanism, the candidate resource sets were significantly reduced to 8 trailers, 2 ISO tanks, 8 vehicles, and 25 drivers. Consequently, the number of feasible combinations decreased to 3200. This reduction corresponds to a search space decrease in more than five orders of magnitude. These results demonstrate that the rule-based filtering mechanism effectively limits the combinatorial explosion of the solution space, allowing the expert system to generate feasible dispatch alternatives efficiently for practical operational planning scenarios.

Detailed information on the resources to be used for planning after the ES application is provided in

Table 9. Overview of Remaining Vehicles Post Expert System Implementation. *** indicates masked digits in license plate numbers to protect confidentiality.

Number	Vehicle License Plate	Status	Last 1 Month km	Last 3 Months km	Fuel Efficiency	Model Year	Current City
1	34***057	Idle/Available	0	0	0.33072279	2020	Gebze, Kocaeli, Turkey
2	34***938	Idle/Available	0	2653	0.33995724	2020	Gebze, Kocaeli, Turkey
3	34***881	Idle/Available	0	0	0.33559452	2018	Gebze, Kocaeli, Turkey
4	34***618	Idle/Available	0	0	0.35519536	2018	Gebze, Kocaeli, Turkey
5	34***416	Idle/Available	0	0	0.3587721	2018	Gebze, Kocaeli, Turkey
6	34***685	Idle/Available	0	0	0.30928028	2018	Gebze, Kocaeli, Turkey
7	34***877	Idle/Available	3205	13211	0.31495547	2011	Gebze, Kocaeli, Turkey
8	34***177	Idle/Available	40	97	0.35286215	2011	Gebze, Kocaeli, Turkey

,

Table 10. Overview of Remaining ISO-Tanks Post Expert System Implementation. *** indicates masked digits in tank numbers to protect confidentiality.

Number	Tank Number	Product Group	Attached Trailer	Last 1 Month Position	Last 3 Month Position	Last 1 Month Position (Intra City)	Current City
1	EUR***11-8	TDI	-	14	59	0	Gebze, Kocaeli, Turkey
2	EUR***20-5	TDI	-	13	58	0	Gebze, Kocaeli, Turkey

,

Table 11. Overview of Remaining Trailers Post Expert System Implementation. * indicates masked digits in license plate numbers to protect confidentiality.

Number	License Plate	ADR Status	Current City	Attached Tank	Last 1 Month Position	Last 3 Month Position	Last 1 Month Position (Intra City)
1	34B*****1	ADR	Gebze, Kocaeli, Turkey	EUR***98-8	0	3	0
2	34A*****2	ADR	Gebze, Kocaeli, Turkey		0	2	0
3	34H*****9	ADR	Gebze, Kocaeli, Turkey	EUR***08-0	0	6	0
4	34R*****3	ADR	Gebze, Kocaeli, Turkey		0	44	0
5	34R*****5	ADR	Gebze, Kocaeli, Turkey		0	67	0
6	34R*****0	ADR	Gebze, Kocaeli, Turkey		0	15	0
7	34R*****2	ADR	Gebze, Kocaeli, Turkey		0	39	0
8	06A*****6	ADR	Gebze, Kocaeli, Turkey		0	0	0

and

Table 12. Overview of Remaining Drivers Post Expert System Implementation. * indicates masked characters in driver names to protect personal data privacy.

	Driver Name	Src_5 License	Current City	Last 1 Month km	Short Distance	Long Distance
1	Ne*** O***	Yes	Izmit, Kocaeli, Turkey	831	26	3
2	Ne*** Bi***	Yes	Izmit, Kocaeli, Turkey	3642	9	19
3	Re*** S***	Yes	Izmit, Kocaeli, Turkey	326	57	0
4	Se*** Yu***	Yes	Izmit, Kocaeli, Turkey	3426	11	14
5	Mu*** Ba***	Yes	Izmit, Kocaeli, Turkey	4763	13	20
6	Ze*** Kı***	Yes	Izmit, Kocaeli, Turkey	1992	17	12
7	Ha*** Cel***	Yes	Izmit, Kocaeli, Turkey	2107	15	6
8	Tu*** Ko***	Yes	Izmit, Kocaeli, Turkey	5015	14	14
9	Yu*** Ba***	Yes	Izmit, Kocaeli, Turkey	3953	25	20
10	Ra*** Gu***	Yes	Izmit, Kocaeli, Turkey	5087	16	20
11	Se*** Mu***	Yes	Izmit, Kocaeli, Turkey	3803	8	13
12	Ta*** Ca***	Yes	Izmit, Kocaeli, Turkey	5177	19	22
13	Re*** Er***	Yes	Izmit, Kocaeli, Turkey	3456	20	8
14	Ta*** Gu***	Yes	Izmit, Kocaeli, Turkey	4718	6	22
15	Bi*** Yi***	Yes	Izmit, Kocaeli, Turkey	1116	5	3
16	Se*** Bi***	Yes	Izmit, Kocaeli, Turkey	4826	13	22
17	Ol*** Ko***	Yes	Izmit, Kocaeli, Turkey	550	8	2
18	Ha*** Gu***	Yes	Izmit, Kocaeli, Turkey	4316	29	10
19	Re*** Ay***	Yes	Izmit, Kocaeli, Turkey	4987	38	9
20	Me*** Di***	Yes	Izmit, Kocaeli, Turkey	2126	59	0
21	Mu*** Ay***	Yes	Izmit, Kocaeli, Turkey	4714	16	24
22	Er*** Si***	Yes	Izmit, Kocaeli, Turkey	5153	29	15
23	Zi*** Ki***	Yes	Izmit, Kocaeli, Turkey	4415	11	18
24	Se*** Oz***	Yes	Izmit, Kocaeli, Turkey	4908	14	15
25	Ha*** Az***	Yes	Cerkezkoy, Tekirdag, Turkey	4663	18	14

.

According to Table 9, there are eight vehicles, all located in Kocaeli. All the vehicles in the table are marked as “Available,” indicating that they are not currently in service and are available for new assignments. The majority of the vehicles have not traveled any distance in the past month, with only two vehicles having covered some distance (2 and 7). Notably, Vehicle-7 has traveled the longest distance in the last three months, reaching 13,211 km. The fuel consumption rates of the vehicles range from 0.30928028 to 0.3587721. The vehicle with the lowest fuel consumption rate is Vehicle-6. The vehicles have model years ranging from 2011 to 2020. The newest models among these vehicles are Vehicle-1 and Vehicle-2, both from the year 2020.

Table 10 contains information about two tank numbers located in Gebze, Kocaeli. These tanks belong to the TDI product group. It is observed that the tanks have changed position 14 and 13 times in the last month and 59 and 58 times in the last three months, respectively. This indicates that the tanks are actively used and regularly moved. There has been no intra-city position change in the past month, and currently, they are not attached to any trailer.

According to Table 11, a total of 8 trailers located in Kocaeli can be included in the planning following the ES application. All the trailers in the table comply with ADR standards. None of the trailers has undergone any position changes within the city in the past month. However, when examining the data from the last three months, it is evident that some trailers have experienced significant position changes. For example, the Trailer-5 has changed position 67 times in the past three months, indicating heavy usage. On the other hand, some trailers, such as Trailer-1, have undergone fewer position changes. The table also lists the tanks to which some trailers are attached; for instance, Trailer-1 is connected to tank EUR***98-8.

According to Table 12, a total of 25 drivers with an SRC 5 certificate can be included in the planning after the ES application. The driver with the least distance traveled in the past month is Driver-3, who has covered only 326 km. This low mileage suggests that Driver-3 has been less active or has taken fewer assignments compared to other drivers. Driver-17, on the other hand, has traveled 550 km, which is slightly more than Driver-3, but he is still one of the drivers with the lowest mileage among those listed in the table.

The most active driver in short-distance trips is Driver-20, who completed 59 short-distance journeys. This indicates that he is primarily engaged in short-distance tasks. In contrast, Driver-15, who has completed only five short-distance trips, stands out as the driver with the fewest short-distance assignments.

In long-distance trips, Driver-21 leads with 24 long-distance journeys, showing that he is highly active in long-distance transportation. Driver-12 and Driver-16 are also active in long-distance trips, each completing 22 long-distance journeys. On the other hand, Driver-3 and Driver-20 have not undertaken any long-distance journeys.

The resources obtained after the ES were subjected to Cartesian products, and each was scored. The top 10 results from the calculations were presented to the user. The data presented is shown in

Table 13. Planning Recommendations for the Order Following Expert System Application.

Plan Number	ISO Tank Number	Trailer Number	Vehicle Number	Driver Number
1	2	1	1	3
2	2	2	1	3
3	2	3	1	3
4	2	4	1	3
5	2	5	1	3
6	2	6	1	3
7	2	7	1	3
8	2	8	1	3
9	1	1	1	3
10	1	2	1	3

.

According to Table 13, Driver-3 is among the top 10 recommendations. When examining the recommended driver, it is noted that he has traveled the least distance in the past month and has not undertaken any long-distance trips. Since the order is for a long-distance route, Driver-3 is deemed the most suitable driver based on the short/long-distance criteria. The reason for his selection for long-distance transportation is his greater experience with short-distance trips, and by assigning him to long-distance trips, a more balanced workload distribution is aimed at.

Another selected resource is Vehicle-1. This vehicle has not traveled any distance in the past month, making it relatively new and less worn compared to other vehicles. It may also have been selected due to its more efficient fuel consumption. In terms of fuel efficiency, it is observed that Vehicle-1 is the third most fuel-efficient, which provides a significant advantage in reducing operational costs for long-distance transportation. Additionally, it is the newest model among the available vehicles.

4.3. Comparison of Expert System Results

A comparative evaluation was conducted between manual planning performed by the planner and the planning outcomes generated by the expert system to assess the practical feasibility and consistency of the proposed approach. This comparison is presented in

Table 14. Expert System Elimination Details for Manually Selected Resources Across 8 Orders. * indicates masked digits/characters in license plate numbers, tank numbers, and driver names to protect confidentiality and personal data privacy.

Order Number		Manuel	ES
			Rule Number	Description
1	ISO Tank	EUR***50-3	13	Position Count: 15 Average Position Count: 3.833
	Trailer	34H***91	17, 18	Intercity Position Count: 15 60% Intercity Position: 9
	Vehicle	34E***61	19	km: 3709 Average km: 3246.288
	Driver	Tu*** Ko***		Included in the planning
2	ISO Tank	DOV***25-5	14, 15	Intercity Position Count: 5 60% Intercity Position: 3
	Trailer	34B***63	17, 18	Intercity Position Count: 6 60% Intercity Position: 3
	Vehicle	34B***46	10, 11, 12	Ordered City: Kocaeli The city where the vehicle is located: İzmir
	Driver	Ni*** As***	26	km: 5701 Average km = 5248.118
3	ISO Tank	EUR***11-8	-	Included in the planning
	Trailer	34H***45	17, 18	Intercity Position Count: 14 60% Intercity Position: 8
	Vehicle	34B***71	19	km: 5177 Average km: 3475.830
	Driver	Ta*** Ça***	-	Included in the planning
4	ISO Tank	DOV***50-2	13	Position Count: 17 Average Position Count: 14.766
	Trailer	34R***82	16	Position Count: 22 Average Position Count: 17.571
	Vehicle	34B***87	10, 11, 12	Ordered City: Kocaeli The city where the vehicle is located: Tekirdağ
	Driver	Ru*** Tu***	26	km: 6133 Average km = 3800.441
5	ISO Tank	ASN***03-0	13	Position Count: 27 Average Position Count: 26.333
	Trailer	34R***08	16	Position Count: 27 Average Position Count: 23
	Vehicle	34E***62	19	km: 5059 Average km: 3783
	Driver	Şe*** Mü***	-	Included in the planning
6	ISO Tank	DOV***46-2	-	Included in the planning
	Trailer	34L***16	10, 11, 12	Ordered City: Tekirdağ The city where the trailer is located: Kocaeli
	Vehicle	34C***89	10, 11, 12	Ordered City: Tekirdağ The city where the vehicle is located: Kocaeli
	Driver	Se*** Öz***	-	Included in the planning
7	ISO Tank	EUR***00-9	13	Position Count: 37 Average Position Count: 14.766
	Trailer	34R***79	16	Position Count: 37 Average Position Count: 16.5125
	Vehicle	34H***68	10, 11, 12	Ordered City: Kocaeli The city where the vehicle is located: Tekirdağ
	Driver	Ha*** Az***	26	km: 4663 Average km = 4301.3833
8	ISO Tank	DOV***71-3	13	Position Count: 39 Average Position Count: 27.333
	Trailer	34F***26	-	Included in the planning
	Vehicle	34E***53	19	km: 6282 Average km: 5508
	Driver	Ca*** Uy***	26	km: 6282 Average km = 5252.770

.

In Table 14, orders 1, 2, 3, 4, 5, 6, 7, and 8 have been analyzed in both manual planning and the evaluations performed by the ES.

For Order-1, the ISO Tank EUR***50-3 was selected in the manual planning. However, the ES identified that this tank had changed positions 15 times in the last month, which is significantly higher than the average position count of 3.833, and eliminated it by applying Rule-13. This reflects the strategy of prioritizing less-used tanks to ensure balanced use of ISO Tanks. For the same order, the trailer with plate number 34H***91 was eliminated based on Rules 17 and 18, as its intercity position count was 15, exceeding the 60% intercity position criterion of 9. The vehicle with plate number 34E***61 was also eliminated by Rule-19 because the distance it traveled (3709 km) was higher than the average distance (3246.288 km). The driver Tu*** Ko*** was included in the planning without any rule application.

For Order-2, the ISO Tank DOV***25-5 was selected manually, but the ES eliminated it by applying Rules 14 and 15 because its intercity position count was 5, exceeding the 60% intercity position criterion of 3. Similarly, the trailer with plate number 34B***63 was eliminated based on Rules 17 and 18 because its intercity position count was 6. The selected vehicle, 34B***46, was eliminated by Rules 10, 11, and 12 as it was located in İzmir, not in the order’s city of Kocaeli. The driver Ni*** As*** was eliminated by Rule-26 because the distance he traveled (5701 km) exceeded the average distance (5248.118 km).

For Order-3, the manually selected ISO Tank 1 was included in the planning by the ES without applying any rules. However, the trailer with plate number 34H***45 was eliminated by Rules 17 and 18 because its intercity position count of 14 exceeded the 60% intercity position criterion of 8. The vehicle with plate number 34B***71 was eliminated by Rule 19 as the distance it traveled (5177 km) exceeded the average distance (3475.830 km). The driver Ta*** Ça*** was included in the planning without any rule application.

For Order-4, the ISO Tank DOV***50-2 was selected manually, but the ES eliminated it by applying Rule 13 because its position count (17) was higher than the average position count (14.766). The trailer with plate number 34R***82 was eliminated by Rule 16 as its position count (22) exceeded the average position count (17.571). The vehicle with plate number 34B***87 was eliminated by Rules 10, 11, and 12 as it was located in Tekirdağ, not Kocaeli. The driver Ru*** Tu*** was eliminated by Rule 26 because the distance he traveled (6133 km) was higher than the average distance (3800.441 km).

For Order-5, the ISO Tank ASN***03-0 was selected manually, but the ES eliminated it by applying Rule 13 because its position count (27) was close to the average position count (26.333). The trailer with plate number 34R***08 was eliminated by Rule 16 as its position count (27) exceeded the average of 23. The vehicle with plate number 34E***62 was eliminated by Rule 19 because the distance it traveled (5059 km) exceeded the average distance (3783 km). The driver Şe*** Mü*** was included in the planning without any rule application.

For Order-6, the manually selected ISO Tank DOV***46-2 was included in the planning without any elimination by the ES. The trailer with plate number 34L***16 was eliminated based on Rules 10, 11, and 12, as the order was for Tekirdağ, but the trailer was located in Kocaeli, making it unfit for the assignment. Similarly, the vehicle with plate number 34C***89 was also eliminated by the same rules, as it was also located in Kocaeli rather than Tekirdağ. The driver Se*** Öz*** was included in the planning without any rule application.

For Order-7, the ISO Tank EUR***00-9 was manually selected but eliminated by the ES under Rule 13 because its position count (37) was higher than the average position count (14.766). The trailer with plate number 34R***79 was eliminated by Rule 16 because its position count (37) was much higher than the average position count of 16.5125, indicating heavy usage and potentially more wear. The vehicle with plate number 34H***68 was eliminated by Rules 10, 11, and 12 as it was located in Tekirdağ, while the order was for Kocaeli. The driver Ha*** Az*** was eliminated under Rule 26 because the distance he traveled (4663 km) exceeded the average distance (4301.3833 km).

Finally, for Order-8, the ISO Tank DOV***71-3 was manually selected but eliminated by the ES using Rule 13 due to its high position count (39), which surpassed the average position count of 27.333. The trailer with plate number 34F***26 was included in the planning without elimination. The vehicle with plate number 34E***53 was eliminated by the ES under Rule 19 because its traveled distance (6282 km) exceeded the average distance (5508 km). The driver Ca*** Uy*** was eliminated under Rule 26 because the distance he traveled (6282 km) exceeded the average distance (5252.770 km).

The penalty points determined by manual planning and the ES are compared in

Table 15. Comparison of Manual and Expert System Penalty Points Across 8 Orders.

Order Code	Manual Penalty Points	ES Penalty Points
1	0.03790	0.00973
2	0.03651	0.00973
3	0.04414	0.01619
4	0.04615	0.00792
5	0.05058	0.03240
6	0.04319	0.03187
7	0.05976	0.00792
8	0.06621	0.02866

. These penalty points serve as an indicator of how well the selected resources comply with the defined planning criteria and constraints for each order.

An analysis of the data presented in Table 15 reveals that the penalty points assigned by the ES are consistently lower than those calculated through manual planning. For example, for Order-1, the manual penalty point is 0.0379, while the ES’s penalty point is 0.00973. Similarly, in Order-2 and Order-3, it is also evident that the ES’s penalty points are lower than those of the manual system.

The consistently lower penalty points obtained by the expert system indicate that the proposed rule-based approach applies planning criteria in a more consistent and systematic manner compared to manual planning. This result demonstrates the feasibility of the expert system in supporting resource selection decisions under operational constraints. Particularly in Order-5 and Order-6, the smaller difference between the manual and ES penalty points shows that the evaluations of both systems are closer in these cases, and the expert system produces decisions that are closely aligned with manual planning in these cases, indicating compatibility with planner reasoning rather than unconditional replacement. However, in some orders, such as Order-8, the manual penalty point (0.06621) is quite high, while the ES’s penalty point (0.02866) is lower. This emphasizes that the ES provides a more balanced plan.

Table 16. Comparison of manual planning and expert system results.

Order	Method	Fuel	Vehicle km	Driver Workload	Tank Position	Trailer Position
1	ES	0.33	0	5.5	0	0
	Manual	0.32	3709	230.82	15	15
2	ES	0.33	0	5.5	0	0
	Manual	0.36	5701	275.1	5	6
3	ES	0.37	0	5.5	24	1
	Manual	0.27	4789	71.77	0	0
4	ES	0.33	0	5.5	0	0
	Manual	0.34	830	153.67	14	14
5	ES	0.33	0	5.5	0	0
	Manual	0.34	0	241.02	19	20
6	ES	0.29	3721	259.17	18	1
	Manual	0.32	4909	245.38	19	7
7	ES	0.41	0	0	0	0
	Manual	0.29	3721	430.73	37	37
8	ES	0.29	3721	259.17	18	1
	Manual	0.40	6282	240.2	39	39

presents a comparison between manual dispatch planning and the proposed expert system using several operational indicators, including fuel consumption rate, vehicle kilometers, driver workload, and resource utilization levels.

It should be noted that the indicators reported in Table 16 represent historical operational metrics calculated over the last one-month period, rather than the distance or fuel consumption associated with the specific transportation order. In particular, the Vehicle km value corresponds to the total distance traveled by the selected vehicle during the previous month. Similarly, Tank Position and Trailer Position represent the number of transportation assignments completed by the corresponding ISO tank and trailer during the last month. The Driver Workload indicator represents the total number of working hours completed by the driver during the same period. Finally, the Fuel value corresponds to the fuel consumption rate of the selected vehicle.

These indicators are used in the proposed expert system as operational balancing metrics. Instead of selecting resources solely based on proximity or immediate availability, the system evaluates the recent utilization history of each resource to prevent excessive use of the same equipment. By considering past vehicle kilometers, tank position counts, trailer position counts, and driver workload levels, the expert system promotes a more balanced distribution of operational workload across the fleet.

The comparison results indicate that the expert system frequently selects resources with lower recent usage levels compared with manual planning. For example, in several orders the expert system assigns vehicles that have traveled fewer kilometers in the previous month, tanks with lower recent position counts, and drivers with lower accumulated working hours. This behavior directly supports the objective of balancing resource utilization across transportation assets.

In contrast, manual planning decisions may sometimes assign resources that have already been heavily utilized during the previous operational period. This may lead to uneven workload distribution among drivers and equipment, potentially increasing operational fatigue, maintenance requirements, and long-term fleet imbalance.

Table 17. Performance improvement of the expert system.

Order	km Reduction (%)	Fuel Reduction (%)	Driver Workload Reduction (%)	Tank Position Reduction (%)	Trailer Position Reduction (%)
1	100.00	−3.13	97.62	100.00	100.00
2	100.00	8.33	98.00	100.00	100.00
3	100.00	−37.04	92.34	-	-
4	100.00	2.94	96.42	100.00	100.00
5	0.00	2.94	97.72	100.00	100.00
6	24.20	9.38	−5.62	5.26	85.71
7	100.00	−41.38	100.00	100.00	100.00
8	40.77	27.50	−7.90	53.85	97.44

summarizes the performance improvements achieved by the proposed expert system across five operational indicators derived from Table 16. The percentage values represent the relative difference between the resources selected by the expert system and those selected during manual dispatch planning. It should be noted that in Order 3, manual planning reports zero values for both tank and trailer position counts; since percentage reduction cannot be computed relative to a zero baseline, no reduction value is reported for these indicators in Order 3.

The results show that the expert system consistently reduces vehicle kilometer values in most orders. Complete reductions (100%) are observed in Orders 1–4 and 7, indicating that the system selected vehicles with significantly lower historical mileage compared with manual planning. Moderate reductions are also observed in Orders 6 and 8, while Order 5 shows no improvement as both methods selected vehicles with similar mileage levels.

Fuel reduction values show a more mixed pattern. Positive fuel savings are observed in Orders 2, 4, 5, 6, and 8. However, negative fuel reduction values occur in Orders 1, 3, and 7, meaning that the vehicles assigned by the expert system operate at slightly higher fuel consumption rates than those selected during manual planning. These negative values arise because the expert system simultaneously evaluates multiple operational criteria—including workload balancing, ADR compliance, and equipment availability—rather than optimizing fuel consumption in isolation. This multi-criteria structure inherently produces trade-offs in individual metrics.

Driver workload results demonstrate significant improvements in most orders, particularly in Orders 1–5 and 7, where the expert system selects drivers with substantially lower accumulated working hours. Negative workload improvements in Orders 6 and 8 indicate that operational constraints such as location compatibility and ADR certification requirements led the system to prioritize constraint adherence over workload minimization in these specific cases.

Similar patterns are observed in tank and trailer utilization. In most orders, the expert system selects equipment with lower recent position counts, supporting a more balanced distribution of equipment usage across the fleet. This effect is particularly visible in Orders 1, 2, 4, 5, and 7. Smaller improvements or minor trade-offs in Orders 6 and 8 reflect the same multi-criteria prioritization described above.

To further validate these observations statistically, a paired t-test was conducted to examine whether the differences between manual dispatch planning and the expert system are statistically significant across the evaluated operational indicators. The results of the paired t-test are presented in

Table 18. Paired t-test results.

Metric	Mean Difference (Manual − ES)	t-Value	p-Value	Interpretation
Vehicle km	2812.38	3.95	0.0055	Significant
Driver Workload	167.86	3.08	0.017	Significant
Tank Position	11	1.75	0.123	Not significant
Trailer Position	16.88	3.34	0.012	Significant

.

The results indicate that the expert system produces statistically significant improvements in vehicle kilometers (p = 0.0055), driver workload (p = 0.0177), and trailer position utilization (p = 0.0124). These findings confirm that the proposed rule-based expert system effectively reduces historical vehicle usage, balances driver workload, and distributes trailer utilization more evenly across the fleet.

In contrast, the difference observed for tank position utilization (p = 0.1245) does not reach statistical significance. This outcome reflects the multi-criteria decision structure of the expert system, which simultaneously considers resource availability, compatibility conditions, and workload balancing constraints. Additionally, the outlier value observed in Order 3 substantially increases variance and reduces statistical power for this metric.

Overall, the statistical analysis confirms that the proposed expert system produces significant improvements across the majority of evaluated operational indicators compared with manual dispatch planning.

4.4. Discussion

The results demonstrate that the proposed expert system substantially reduces the decision search space by sequentially eliminating infeasible resources. For Order-3, the number of candidate trailers was reduced from 214 to 8, ISO tanks from 187 to 2, vehicles from 184 to 8, and drivers from 164 to 25. This level of reduction allows dispatch decisions to be made in a fast and manageable manner under strict operational time constraints, which is particularly important in real-world environments where decisions are often required within seconds rather than minutes.

The comparison between manual planning and expert system evaluations shows that the proposed system does not indiscriminately override planner decisions. In several cases, such as Orders 3, 5, and 6, manually selected resources were retained by the expert system without elimination. This indicates that the rule base reflects expert reasoning patterns commonly applied by planners, while applying the same decision logic consistently across different orders.

Sensitivity analysis was conducted to examine the impact of varying the weight coefficients in the composite scoring function. Different weighting scenarios were designed to emphasize specific operational priorities, such as fuel efficiency, driver workload balance, or equipment utilization. The results indicate that changes in weight values primarily affect the relative ordering of feasible alternatives, while the top-ranked solutions remain largely stable across scenarios. ISO tank, vehicle, and driver assignments were preserved, with only minor variations observed in trailer selection. This behavior demonstrates that the proposed decision framework is robust with respect to reasonable parameter adjustments and allows planners to adapt the system’s focus to operational preferences without compromising feasibility, safety, or consistency.

From an operational perspective, the proposed expert system provides dispatch planners with a transparent and explainable decision-support tool. By enforcing safety, compatibility, and workload balancing rules, the system reduces the likelihood of human oversight and supports regulatory compliance, particularly in hazardous liquid transportation. The results highlight the feasibility of integrating the system into daily dispatch operations to support faster, safer, and more consistent planning decisions.

5. Conclusions

This study presented a rule-based expert system designed to support dispatch planning in liquid transportation operations. The proposed system formalizes the decision logic used by experienced planners and transforms this knowledge into a structured rule-based framework that evaluates feasible combinations of heterogeneous transportation resources, including trailers, ISO tanks, vehicles, and drivers.

The results obtained from real operational data demonstrate that the proposed expert system can effectively support dispatch planning decisions. The rule-based filtering mechanism significantly reduces the combinatorial search space by eliminating infeasible resource combinations before the evaluation stage. In the experimental analysis, the initial solution space exceeding one billion possible combinations was reduced to a manageable number of feasible alternatives, enabling efficient decision generation in operational environments where rapid planning is required.

The experimental comparison with manual dispatch planning indicates that the expert system can improve operational efficiency by selecting resource combinations with lower vehicle kilometers, reduced fuel consumption, and more balanced workload distribution among transportation resources. Furthermore, statistical analysis using a paired t-test confirmed that the reductions in vehicle kilometers and fuel consumption are statistically significant. These findings support the research hypothesis that expert knowledge used in liquid transportation dispatch planning can be systematically formalized into a rule-based decision-support framework capable of generating feasible, consistent, and operationally reliable resource allocation decisions under complex operational constraints.

From a sustainability perspective, improved dispatch planning can also contribute to more efficient use of transportation resources. By avoiding infeasible or inefficient resource combinations, the proposed expert system may help reduce unnecessary vehicle movements and improve operational efficiency in logistics operations. More efficient resource utilization may also indirectly contribute to reduced fuel consumption and lower environmental impacts associated with transportation activities. Although the present study primarily focuses on operational feasibility and decision-support performance, improved dispatch planning has the potential to support sustainability objectives in logistics systems by reducing operational inefficiencies, unnecessary trips, and resource waste.

From an operational perspective, the proposed system functions as a decision-support tool rather than a fully autonomous planning mechanism. By maintaining transparency in the rule-based reasoning process and allowing planners to review and adjust recommendations, the system supports human-centered decision-making while improving consistency and reducing the risk of planning errors.

Considering the limitation that the current evaluation is based on a relatively small dataset obtained from a single logistics company, consisting of eight transportation orders, future research should include larger and more diverse datasets collected from multiple companies and operational contexts. Another limitation of this study is that the proposed rule-based expert system is not directly compared with mathematically optimal solutions obtained from optimization-based models such as mixed-integer programming formulations. Instead, the evaluation focuses on comparisons with manual dispatch planning practices used in real operational environments. Future research may integrate optimization-based solution approaches or hybrid decision-support frameworks to benchmark the performance of rule-based systems and further investigate the optimality gap between heuristic and exact methods. Such studies would enable broader empirical validation of the proposed approach and allow the generalizability of the results to be examined more systematically. In addition, systematic multi-criteria decision-making approaches such as the Fuzzy Analytic Hierarchy Process (FAHP) may be investigated to determine the penalty weights used in the evaluation function in a more structured manner based on expert assessments. Furthermore, future studies may extend the proposed rule-based framework to additional liquid transportation sectors, including petroleum and gas transportation. Incorporating these sectors would require the integration of additional regulatory, safety, and compatibility rules specific to hazardous energy products, which could further enhance the applicability of the expert system in broader industrial logistics contexts.

Overall, the findings indicate that rule-based expert systems provide a practical and scalable framework for supporting complex dispatch planning decisions in liquid transportation environments characterized by strict operational, regulatory, and compatibility constraints.