Application and Optimization of Industrial Internet and Big Data Analytics in Enterprise Decision-Making ^†

Abstract

The integration of the industrial Internet and big data analytics is reshaping enterprise decision-making models and providing new momentum for the transformation and upgrading of traditional manufacturing industries. In this study, a decision support system based on multi-source heterogeneous data fusion was established. The system carries out data collection, storage, and processing, as well as visualization analysis. The system also performs time-series data feature extraction and unstructured data processing in a three-layer architecture model to train models and generate decision-making. In case studies, the effectiveness of the system in predictive maintenance of equipment, dynamic optimization of supply chains, and product quality traceability was verified. A fault prediction model was developed based on an improved random forest algorithm, and it showed a high level of accuracy. Optimization strategies, such as modular system design, dynamic knowledge base updating, and human–machine collaborative decision-making, can be formulated using the system. To evaluate the system, a three-dimensional evaluation index system was built, including technology maturity, application adaptability, and benefit–output ratio. The developed system effectively improved the efficiency of enterprise resource allocation, shortened abnormality response times, and enhanced market adaptability. By using edge computing and digital twin technologies, a more flexible distributed decision-making architecture can be created in the system, promoting data-driven and intelligent decision-making in manufacturing industry.

1. Introduction

As important parts of modern information technology, the industrial Internet and big data analytics meet the urgent need for digital transformation of traditional manufacturing. With intensifying competition in the global market, enterprises require efficient resource allocation and precise decision-making support [1].

The industrial Internet enables interconnections between production equipment, supply chain links, and information systems, and hardware such as Internet of Things (IoT) devices and smart sensors. This connectivity enables enterprises to obtain key data concerning equipment operating parameters and material flow status in real time for subsequent analysis [2]. In addition, breakthroughs in big data analytics have led to the development of mature distributed computing frameworks and machine learning algorithms to process massive quantities of heterogeneous data [3]. This technological advancement forms a complete closed loop of “data collection–transmission storage–intelligent analysis”

The synergistic effect of the industrial Internet and big data analytics is manifested in three ways in device-level data collection and cloud-based analysis. First, enterprises can overcome the limitations of traditional empirical decision-making [4]. For instance, an automotive parts manufacturer reduced equipment downtime by 40% through the real-time monitoring of production line data. Secondly, multi-source data fusion analysis enhances the global perspective of decision-making. A home appliance enterprise successfully improved inventory turnover by 25% after integrating supplier data, market feedback, and production records. Lastly, continuous updating of the dynamic knowledge base endows the decision-making system with self-optimization capabilities. An equipment manufacturing enterprise increased maintenance response speed by 30% by establishing a fault feature database.

The current development presents two characteristics. On the one hand, the application of 5G networks and edge computing is transforming data analysis from a centralized to a distributed architecture. One smart factory has achieved millisecond-level data processing latency through the deployment of edge nodes [5]. On the other hand, digital twin technology enables enterprises to make decisions in a virtual environment. One aircraft manufacturer has shortened the verification cycle for new processes by 60% by constructing a digital twin model. Such technological advancements are reshaping the ways of enterprise decision-making, enabling the intelligent transformation of the manufacturing industry [6].

2. Technical Foundations of Industrial Internet and Big Data Analytics

The industrial Internet architecture consists of a perception layer, a network layer, a platform layer, and an application layer in a complete closed loop of data flow. The perception layer collects data from the physical world through devices such as smart sensors and RFID tags. Related systems comprise three core elements: first, an IoT communication protocol, where different protocols exhibit differences in transmission distance, power consumption, and other indicators (

Table 1. Comparison of typical IoT communication protocols.

Protocol	Distance	Power Consumption Level	Typical Application Scenarios
MQTT	Medium-to-long distance	Low	Remote monitoring of equipment status
CoAP	Mid-range	Lower	Intelligent warehouse management system
LoRaWAN	Long distance	Very low	Wide-area environmental monitoring
NB-IoT	Long distance	Medium	Urban infrastructure monitoring

); second, edge computing, with which computer numerical control (CNC) machine tools can adjust the operation according to real-time machining accuracy by deploying edge nodes; third, data security, as when a chemical enterprise employs blockchain technology to ensure the tamper-proof transmission of process parameters. These technologies are used together to ensure the stable operation of industrial Internet systems [7].

For protocol selection, the deployment environment and business requirements need to be considered. In making the selection, technological advancements such as the industrial Internet are used [8].

The processing algorithms and frameworks of big data analytics constitute the technical backbone of enterprise decision-making systems. Classification, clustering, and regression algorithms are widely used to meet the needs of different business scenarios. Classification algorithms categorize data through feature matching. For example, in equipment fault diagnosis, support vector machines (SVMs) distinguish between normal and abnormal vibration signals. Clustering algorithms are used for customer segmentation, and the K-means algorithm automatically groups users based on their consumption behavior. Regression algorithms play an important role in supply chain demand forecasting, and linear regression models are used to explore correlations between inventory levels and sales cycles.

The advancement in distributed computing frameworks has significantly improved data processing efficiency (

Table 2. Comparison of computing framework characteristics.

Characteristic	MapReduce	Spark
Data processing mode	Batch processing	Batch processing/stream processing
Memory utilization	Disk storage is the primary method	Memory priority
Iterative support	Multiple reads and writes to the disk are required	Memory cache intermediate results

). MapReduce uses a divide-and-conquer strategy to carry out batch processing tasks that involve massive amounts of data, partitioning data in the Map phase, and aggregating processing in the Reduce phase. Its computation time is calculated as follows:

T_mapreduce = N × (T_read + T_map + T_shuffle + T_reduce)

(1)

where N represents the number of iterations, and T_shuffle denotes the time taken for data shuffling.

The Spark framework optimizes the iteration process through in-memory computing. Its computational model is as follows:

T_spark = T_read + K × (T_compute) + T_write

(2)

where K represents the number of iterations in memory, eliminating time loss caused by repeated instances of disk-in and disk-out. The Spark framework has advantages in machine learning tasks that require multiple iterations [9].

In industrial settings, combined algorithms and frameworks are used under specific principles. For time-series data analysis, a sliding window mechanism combined with long short-term memory (LSTM) neural networks may be employed. Unstructured data processing is also employed for natural language processing using the term frequency–inverse document frequency (TF-IDF) algorithm, which extracts key fault features from maintenance records. With the development of graph computing frameworks, the complex relationships involved in in supply chain network optimization can be explored for technical support.

3. Application of Enterprise Decision Support Systems

In a production system, real-time data analysis is necessary to provide a basis for production process optimization and dynamic decision-making through the continuous monitoring of equipment status and process parameters [10]. The system needs operational data, including vibration, temperature, and current, which are collected from a sensor network, and combined in a manufacturing execution system (MES) for work order execution in a multi-dimensional data stream. After preliminary cleaning by edge computing nodes, the data are transmitted to the cloud platform for deep processing [11].

Overall equipment effectiveness (OEE) is a core evaluation metric which is calculated based on the three factors of availability, performance, and yield, as follows:

OEE = Availability × Performance × Yield

(3)

The availability rate reflects the equipment time utilization rate, calculated by the ratio of actual operating time to planned working hours; the performance rate reflects the equipment speed efficiency, determined by the product of theoretical cycle time and actual output; and the yield rate is calculated to measure the production quality level.

This model is used to automatically collect real-time data and dynamically generate a heat map of production line efficiency, helping managers quickly identify bottleneck processes (

Table 3. Sources of OEE data.

Data Dimension	Collecting Device	Calculation Elements
Running duration	Plc controller	Availability
Tact time	Photoelectric sensor	Performance rate
Quantity of defective products	Visual inspection system	YIELD
Energy consumption data	Smart meter	Energy efficiency assessment

).

In the electronics assembly industry, a material loss warning mechanism is used to analyze the data on the material ejection rate of the pick-and-place machine, combined with the vibration characteristics of the feeder. When abnormal vibrations at a specific frequency are detected, the system triggers the feeder calibration procedure, significantly reducing material waste. At the same time, the dynamic scheduling module automatically adjusts the work order according to the equipment load status and real-time production capacity, improving the balance of the production line.

The implementation of such a system provides three optimization effects. Firstly, response times in cases of equipment anomalies are shortened from several hours of traditional manual inspection to minutes. Secondly, through real-time feedback adjustment of process parameters, production capacity in key processes is significantly improved and stabilized. Finally, through automatic correlation analysis of quality traceability data, the efficiency of root-cause identification in cases of product defects is considerably enhanced. These improvements verify the value of real-time data analysis in the optimization of production processes [12].

The prediction and simulation of supply chain decisions involves integrating historical data and real-time information to construct a dynamic analysis model that supports enterprises in making precise supply plans. The model collects multi-dimensional data sources, including historical sales records, market trend indicators, supplier delivery cycles, and logistics network status, and forms a data pool for the entire supply chain. After cleaning and feature extraction, the data are input into the prediction model for demand analysis.

In the demand analysis, the autoregressive integrated moving average (ARIMA) model is widely used due to its good adaptability to time series data. This model consists of autoregressive terms (AR), differencing terms (I), and moving average terms (MA), which are applied in the following mathematical expression:

(1 − φ₁B − … − φₚBᵖ)(1 − B)ᵈYₜ = (1 + θ₁B + … + θ_qB^q)εₜ

(4)

where p is the autoregressive order, d is the number of differences, q is the moving average order, B is the lag operator, and εₜ is the white noise sequence.

Through parameter fine-tuning, the model effectively captures the seasonal fluctuations and trend characteristics of sales data (

Table 4. Typical input features of supply chain prediction models.

Feature Category	Sample Data	Acquisition Frequency
Market demand	Historical sales, promotional activities	Daily/weekly
Supply capacity	Supplier on-time delivery rate	Real-time update
External environment	Raw materials price index	Monthly
Logistics status	Transportation route congestion warning	Monitoring

).

Simulation technology simulates the implementation effects of different decision-making schemes using digital twin models. The supply chain simulation system is used to dynamically assess the impact of sudden order growth on inventory turnover and automatically generate multi-level replenishment strategies. When delays are detected at regional logistics nodes, the system immediately generates alternative transportation plans, significantly shortening response times in cases of anomalies.

The system enhances demand-forecasting accuracy and enables reasonable control of safety stock levels. The supplier evaluation model optimizes the partner selection mechanism by quantifying delivery stability indicators. The logistics path simulation module identifies the efficient configuration of the distribution network by calculating the balance between transportation cost and timeliness. These improvements verify the practical value of prediction and simulation technology in supply chain decision-making.

4. Research on Optimization Strategies for Decision Support Systems

The optimization of decision-making models with multi-source data fusion requires addressing issues such as diverse data sources and complex structures. Data generated by equipment sensors, production logs, supply chain systems, etc., have different formats and characteristics. Traditional analysis methods based on a single data source are no longer available for precise decision-making. Therefore, a phased optimization strategy is used to enhance model performance through feature selection and data fusion.

In data preprocessing, a unified data cleaning rule library is established. For noise interference in equipment vibration signals, sliding window mean filtering is employed for smoothing processing. For text-based maintenance records, key entities, such as fault locations and handling measures, are extracted through natural language processing techniques.

For feature selection, the entropy method is employed to screen key indicators. This method is used to evaluate the discriminative ability of features by calculating their information entropy, using the following mathematical expression:

H(X) = −Σ(p(x_i)log₂p(x_i))

(5)

where p(x_i) represents the probability of feature X taking the ith value.

Taking machine tool fault prediction as an example, entropy calculations are performed on 20 monitoring parameters such as temperature, current, and vibration frequency. Eight features with information entropy below the threshold are selected, reducing computational complexity while ensuring prediction accuracy.

In the model architecture optimization, a hierarchical fusion strategy is used. The bottom layer processes structured data, such as equipment operating parameters, the middle layer integrates semi-structured data, such as work order records, and the top layer fuses unstructured data, such as quality inspection images.

In the implementation process, three key points need to be noted: first, a feature engineering knowledge base is established to store the optimal feature combinations in different scenarios; second, a visual feature correlation analysis tool is developed to understand the interactions between data; third, a model iteration triggering mechanism is set up to automatically initiate the retraining process when there is a significant shift in data distribution. These measures are collectively used to ensure the adaptability and reliability of the decision-making model in complex industrial environments.

In the industrial Internet environment, the architecture of edge computing and cloud platforms effectively balances real-time responses and deep analysis by using a hierarchical processing mechanism (

Table 5. Comparison of latency characteristics of two calculation modes.

Evaluation Dimension	Edge Computing Mode	Traditional Cloud Computing Model
Data processing latency	Very low	Higher
Data transmission distance	Local closed-loop system	Remote transmission
Internet dependency	Weak	Strong
Complex computation support	Fundamental analysis	Deep modeling
Energy consumption level	Lower	Higher

). This architecture consists of edge nodes, regional gateways, and cloud centers for data processing with a “device–edge–cloud” linkage. Edge nodes are deployed on the device side and are responsible for real-time data filtering and preliminary analysis. Regional gateways are employed to integrate workshop-level data streams and perform complex event processing, while cloud centers carry out cross-departmental data fusion and model training.

5. Conclusions

Use of the industrial Internet and big data analytics in enterprise decision-making enables reproducible experiences for the transformation and upgrading of manufacturing industries. In different case studies, it was validated that modular system design effectively reduces deployment difficulty; that the continuous updating mechanism of the dynamic knowledge base ensures the adaptability of decision-making models; and that the human–machine collaborative decision-making mode is effective in complex scenarios. The system integrates the experience-based judgments of operators to recommend solutions.

Technological convergence is widely observed today. The integration of edge computing and digital twin technology enables on-site decision-making. Blockchain technology bolsters data credibility, ensuring the immutability of product lifecycle data. Augmented analytics technology lowers the threshold for data analysis, enabling ordinary managers to obtain decision-making recommendations through natural language interaction.

In applications, decision support systems improve the intelligence of production processes by embedding self-learning algorithms for the optimization of process parameters. These are applied in the industrial chain with a mechanism for data sharing between enterprises. By sharing data between upstream and downstream enterprises, inventory turnover efficiency can be enhanced. At the same time, the decision evaluation system needs to be upgraded for system health diagnostics. It is recommended that evaluation indicators be constructed from three dimensions: technical maturity, application adaptability, and benefit–output.

Ethical issues must be seriously addressed, especially with regard to data ownership and algorithm transparency. Enterprises need to establish data classification and authorization systems to manage safe and controllable data circulation models while ensuring core competitiveness. Balancing technological innovation and risk management ensures the healthy development of the industry.