Refuse-Derived Fuel with the Addition of Peanut Shells: An Evaluation Using a Decision-Making Support Algorithm

Abstract

Brazil has made progress in Municipal Solid Waste (MSW) management through national legislation focused on integrated waste handling. However, challenges persist, particularly regarding MSW overproduction. A sustainable alternative is Refuse-Derived Fuel (RDF), generated from MSW with or without biomass addition. To be viable for combustion, RDF must meet established energy and environmental quality standards. In this context, a mathematical model based on fuzzy logic was developed to classify RDF quality and support decision-making. Five RDF samples were tested, evaluating their Lower Heating Value (LHV), chlorine, and mercury contents using calorimetry, atomic absorption, and X-ray fluorescence. Results indicate that RDF produced solely from MSW tends to have inadequate LHV, necessitating drying pretreatment. Even with the addition of peanut shells, the highest classification achieved was “Regular”, suggesting limited suitability for combustion in furnaces or boilers without pretreatment. Since the general composition of MSW in Brazil is consistent with the characteristics analyzed, RDF may remain unviable for energy recovery under similar conditions. Economic feasibility studies on drying are recommended, especially in urban centers with limited landfill space.

1. Introduction

One of the major global challenges today is the management of Municipal Solid Waste (MSW), whose generation has significantly increased in recent years. According to the Global Waste Management Outlook 2024, published by the United Nations Environment Programme (UNEP) [1] in partnership with the International Solid Waste Association (ISWA), global MSW generation was estimated at 2.3 billion tons in 2023. Projections indicate that, if current trends continue, this volume could reach 3.8 billion tons by 2050, representing an increase of approximately 65%. Additionally, the report highlights that global costs associated with waste management—including environmental and public health impacts—could nearly double by 2050, reaching USD 640.3 billion annually.

MSW is generated through household activities and the cleaning of public areas [2]. Although often considered waste, these materials can be repurposed. In developing economies such as Brazil, China, and India, MSW generation tends to grow more rapidly due to accelerated urbanization, population growth, and economic development [3]. In Brazil alone, the National Water and Sanitation Agency (NWSA) reported that over 65 million tons of MSW were collected in urban areas in 2021 [4], increasing to approximately 81.8 million tons in 2022—equivalent to 224 thousand tons per day [5].

Despite advancements in MSW management, particularly following the implementation of the National Policy on Solid Waste (NPSW), Law No. 12.305 [2], Brazil continues to face significant challenges. These include reducing the environmental impacts of waste and addressing social issues, such as integrating informal waste collectors into cooperatives and associations.

A key obstacle remains the improper final disposal of large volumes of MSW. Official data from the NPSW indicated that, in 2021, 9.6 million tons of MSW were dumped in open landfills, while 7.55 million tons were disposed of in controlled landfills [6]. Data from the Brazilian Association of Public Cleaning and Special Waste Companies (ABRELPE) [5] confirm that 39% of waste in Brazil was still disposed of inappropriately in 2022. Therefore, further efforts are needed to improve final disposal practices to mitigate environmental damage and safeguard public health [7].

According to the World Health Organization (WHO) [8], every dollar invested in basic sanitation—including waste management—can result in savings of four to five dollars in healthcare costs. Thus, investing in waste management not only improves public health but also enhances the efficiency of public spending.

One alternative for the treatment and final disposal of MSW is its conversion into Refuse-Derived Fuel (RDF) for energy recovery. RDF can be used to produce steam, electricity, or hot water. Electricity can be fed into the grid and distributed to consumers, hot water can supply district heating and cooling systems, and steam can be utilized in nearby industrial processes [9].

However, for waste combustion to be viable, the RDF must have a Lower Heating Value (LHV) above 2390 kcal∙kg⁻¹ [1]. In countries with average MSW characteristics, such as Brazil, the LHV typically ranges between 1100 and 1300 kcal∙kg⁻¹ due to high moisture content [10].

A practical solution to reduce humidity and increase LHV is the incorporation of drier residues, such as agricultural biomass. Managing the disposal of these agro-industrial byproducts is also a growing challenge, as substantial quantities are routinely generated.

The Food and Agriculture Organization (FAO) [11] estimates that around 1.3 billion tons of agro-industrial waste are produced annually, much of which results from processing across the production chain. These byproducts can be repurposed for energy generation in industrial furnaces, such as in sugar mills using sugarcane bagasse.

Agro-industrial biomass offers a sustainable alternative to fossil fuels in energy production [12]. Among various biomasses, peanut shells stand out for their high LHV of 3164.77 kcal·kg⁻¹ [13]. Recent studies reinforce the relevance of biomass in co-combustion scenarios involving waste and renewable inputs. For example, the feasibility of biomass-based waste-to-energy routes in the electricity sector has been evaluated, highlighting the role of biomass in reducing carbon emissions and enhancing the efficiency of thermal processes [14]. Such findings support the strategic use of agro-industrial byproducts, like peanut shells, in RDF production and energy recovery applications.

In addition to their energy potential, peanut shells are particularly relevant to Brazil’s agricultural economy, especially in municipalities within the state of São Paulo. In Tupã, the study site, peanut shell availability is significant due to local grain processing industries. According to the Federation of Industries of the State of São Paulo (FIESP) [15], in 2020, Tupã’s peanut export revenue accounted for 20.7% of the state’s total.

Recent advances in the application of artificial intelligence and data-driven methods have significantly contributed to the development of more accurate and automated systems for Refuse-Derived Fuel (RDF) classification. For example, image-based techniques combined with machine learning algorithms—such as random forest classifiers and neural networks—have proven effective in identifying RDF fractions from visual data [16]. Furthermore, convolutional neural networks (CNNs) have been applied in regression models to predict the Higher Heating Value (HHV) of RDF based on spectrographic and elemental information, achieving high predictive accuracy [17]. These techniques complement fuzzy models by offering alternative approaches to address the heterogeneity of municipal solid waste. In addition, recent reviews have emphasized the importance of incorporating sustainable practices into RDF production, particularly in waste processing facilities [18]. Life cycle assessment (LCA) studies, such as one conducted in Korea, further reinforce the significance of evaluating RDF systems from a cradle-to-grave perspective [19]. A broader overview of RDF classifications and their policy relevance is also presented in recent snapshot reviews, highlighting their role in waste-to-energy strategies [20].

Based on this context, the present research addresses two pressing issues: (i) the environmentally appropriate disposal of MSW to reduce landfill use, and (ii) the reuse of peanut shells—a byproduct of agro-industrial processing. To achieve this, it is essential to evaluate how MSW mixed with biomass burns and assess the quality of the resulting RDF compound.

This study seeks to answer the following questions: How does the energy generation process occur from RDF composed solely of MSW? Can RDF be effectively produced by combining MSW with biomass, particularly peanut shells? Which variables are critical for assessing RDF quality? Is it possible to develop a mathematical model to classify RDF quality for use in Brazilian municipalities?

The general objective of this work is to develop a decision-making support algorithm capable of classifying the quality of RDF generated from MSW, with or without biomass addition.

The specific objectives are to (a) conduct a literature review on RDF and the use of biomass, especially peanut shells, as a fuel supplement; (b) identify key characteristics for RDF classification to define input variables for the model; (c) develop a mathematical model using fuzzy logic to classify RDF and assist in decision-making; (d) apply the model in a municipality in the interior of São Paulo to validate its effectiveness.

2. Theory

Environmentally appropriate waste disposal is a complex task and cannot be implemented uniformly across all municipalities, as the particle size and composition of Municipal Solid Waste (MSW) vary depending on the collection location. Therefore, the specific characteristics of waste in each locality must be considered when proposing a system for its final disposal.

Figure 1 presents the percentage of each component of MSW in Brazil, while Figure 2 illustrates the percentage of recyclable materials found in the municipality of Tupã, which was analyzed in this study.

As shown in Figure 1, organic waste—also referred to as the “wet fraction”—is the primary component of MSW in Brazil, accounting for 51.4%. Recyclable waste constitutes 31.9% and primarily consists of plastic (13.5%), paper and cardboard (13.1%), metals (2.9%), and glass (2.4%). The “other” category includes unrecyclable and contaminated waste, as well as miscellaneous unidentified materials.

When waste is subjected to thermochemical conversion processes, characteristics such as moisture content and the Lower Heating Value (LHV) are crucial. The higher the moisture content, the more difficult it is to burn the waste, resulting in a lower heating value.

Table 1. Feasibility of incineration processes according to the LHV of MSW.

LHV	Feasibility	Justification
LHV < 1.675 kcal·kg−1	Technically unfeasible	Requires auxiliary fuel and presents technical challenges.
1.675 kcal·kg−1 < LHV < 2.000 kcal·kg−1	Partially feasible	Requires pre-treatment to increase heating value.
LHV > 2.000 kcal·kg−1	Technically feasible	Suitable for mass burning plants, generating 450 to 700 kWh per ton of MSW.

Source: Adapted from [23] (p. 28).

summarizes the feasibility of incineration processes based on the LHV of MSW.

The data indicate that combining MSW with sugarcane straw and eucalyptus bark for thermal energy production in boiler systems improves fuel quality. The resulting RDF becomes more homogeneous and less humid, and has a higher LHV, while also offering improved storage, transport, and handling properties [24].

In addition to the LHV, Brazilian Standard NBR 16.849/2020 [1] also requires the assessment of chlorine and mercury content for the classification of MSW intended for energy recovery, given the harmful effects of these elements on human health and the environment.

As shown in

Table 2. Limits for MSWEGP classification.

Classification Feature
Unit	Statistical Measure	Classes
LHV (dry sample)		P1	P2	P3
kcal·kg−1	Limit below average (p ≥ 95%)	LHV ≥ 4.750	4.750 > LHV ≥ 3.580	3.580 > LHV ≥ 2.390
Chlorine content		C1	C2	C3
%	Limit above average (p ≥ 95%)	Cl ≤ 0.5	0.5 < Cl ≤ 1.5	1.5 < Cl ≤ 3.0
Mercury content		H1	H2	H3
mg·kg−1	Arithmetic mean	Hg ≤ 0.1	0.1 < Hg ≤ 0.25	0.25 < Hg ≤ 0.5
	Percentile of 80	Hg P80 ≤ 0.2	0.2 < Hg P80 ≤ 0.5	0.5 < Hg P80 ≤ 1

Source: [25].

, a material must have a Lower Heating Value (LHV) of at least 2390 kcal·kg⁻¹ on a dry basis, and must contain no more than 3.0% chlorine and 1.0 mg·kg⁻¹ mercury to be considered an MSWEGP (Municipal Solid Waste for Energy Generation Purposes).

In addition to Municipal Solid Waste for Energy Generation Purposes (MSWEGP), as defined by NBR 16.849/2020 [25], some regulations use the term Refuse-Derived Fuel (RDF). RDFs are classified into seven categories according to the American Society for Testing and Materials (ASTM) [26], based on the physicochemical characteristics of the materials—specifically, their heating value, and chlorine and mercury contents (

Table 3. RDF categories classified according to their physicochemical characteristics.

Category	Description
RDF-1	Waste used for disposal; lower heating value greater than 5.97 kcal·kg−1; chlorine content less than 0.2%; mercury content less than 0.04 mg·MJ−1.
RDF-2	Also known as coarse RDF; these wastes are processed into coarse particle sizes, with or without the separation of ferrous metals, so that 95% by weight passes through a 6-inch square mesh; net heating value greater than 4.78 kcal·kg−1; chlorine content less than 0.6%; mercury content less than 0.06 mg·MJ−1.
RDF-3	Also called fluff RDF; these residues are processed to separate glass, metal, and inorganic materials, then crushed so that 95% by weight passes through a 2-inch square mesh; net heating value greater than 3.58 kcal·kg−1; chlorine content less than 1.0%; mercury content less than 0.16 mg·MJ−1.
RDF-4	Also called powdered RDF; these combustible wastes are transformed into powder, with 95% by weight passing through a 10-mesh screen (0.035 inches); net heating value greater than 2.39 kcal·kg−1; chlorine content less than 1.5%; mercury content less than 0.30 mg·MJ−1.
RDF-5	Also called densified RDF; these combustible wastes are compressed into the form of pellets (rounded and compact pieces), slugs (elongated and typically rounded pieces), blocks, or briquettes; net heating value greater than 717.02 kcal·kg−1; chlorine content less than 3.0%; mercury content less than 1.0 mg·MJ−1.
RDF-6	Combustible wastes converted into liquid fuels.
RDF-7	Combustible wastes converted into gaseous fuels.

Source: Adapted from [27] (p. 113).

) [22].

In industrial applications, the life cycle of Refuse-Derived Fuel (RDF) follows a well-established technological sequence. This sequence includes (i) the collection and transportation of Municipal Solid Waste (MSW) to treatment facilities; (ii) initial sorting, either manually or automatically, to remove recyclables and hazardous or non-combustible rejects; (iii) mechanical and biological treatment, which may involve shredding, screening, magnetic separation, and organic stabilization, in order to reduce moisture content and improve combustion characteristics; (iv) RDF production through densification methods such as granulation or briquetting; (v) storage and transportation of RDF to industrial facilities, such as cement kilns or thermal power plants; (vi) energy recovery through controlled combustion; (vii) proper management of post-combustion residues; and (viii) continuous environmental monitoring throughout the entire cycle. These stages are generally automated and governed by technical standards that ensure efficiency and standardization. Understanding this industrial dynamic is essential to contextualize results obtained at the laboratory scale, where infrastructure limitations may require manual execution of steps such as impurity separation. The review by [20] provides a comprehensive overview of the technological routes for RDF production and the operational and environmental aspects involved in its life cycle.

Next, the methodological procedures used in this study will be presented, including the variables considered, the rule set developed for the fuzzy inference process, and the methodology adopted to estimate the values of the study area to be input into the system.

3. Materials and Methods

In this research, we aimed to develop a model grounded in rigorous scientific studies, enabling its application in different regions of the country to classify RDF based on its quality and in accordance with the standards established by NBR No. 16.849/2020 [25].

3.1. Literature Review on the Proposed Topic

This study included a literature review using global scientific databases (Scopus, Web of Science, ScienceDirect, and SciELO), as well as the CAPES Periodicals Portal, to define what RDF actually is. It also explored the classifications and potential energy applications of refuse-derived fuels.

Additionally, a review of Brazilian and São Paulo state legislation was conducted to identify the regulatory frameworks related to the thermal treatment of waste involving RDF.

3.2. Fuzzy Modeling for RDF Classification

The Fuzzy Logic Toolbox from MATLAB^® R2023b was used to develop membership functions and graphical representations, along with Microsoft Excel spreadsheets. Both software programs are licensed for use at the Faculty of Science and Engineering of São Paulo State University (UNESP).

Based on the literature review conducted in scientific databases and the thresholds established by Brazilian and international legislation, the input variables for the model were defined. It was observed that the three main variables determining RDF quality are lower heating value (LHV), chlorine content, and mercury content. Accordingly, the classes proposed by [1] were adapted, as shown in

Table 4. Limits for RDF classification.

Classification Feature
Unit	Classes
LHV (dry sample) kcal·kg−1	P1	P2	P3	P4
	LHV ≥ 4.750	4.750 > LHV ≥ 3.580	3.580 > LHV ≥ 2.390	LHV < 2.390
Chlorine Content %	C1	C2	C3	C4
	Cl ≤ 0.5	0.5 < Cl ≤ 1.5	1.5 < Cl ≤ 3.0	Cl > 3.0
Mercury Content mg·kg−1	H1	H2	H3	H4
	Hg ≤ 0.1	0.1 < Hg ≤ 0.25	0.25 < Hg ≤ 0.5	Hg > 0.5

P—Lower Heating Value; C—Chlorine Content; H—Mercury Content; Source: Adapted from [25].

. RDF classification was defined as the output variable.

Thus, the following mathematical model was considered to represent the situation: F: A ⊂ IR3 → B ⊂ IR, where R denotes the set of real numbers, I, the proposed input and output variables for the model, A is the set of input variables, and B is the set of output variables (Figure 3).

To establish the set of linguistic rules, the MSWEGP classification limits proposed by [25] were adapted, as shown in Table 4.

Triangular and trapezoidal membership functions were used to define the input variables.

Table 5. Definition of the degree of membership of variables.

Variable	Fuzzy Set	Type of Function	Limiting Factor
Lower Heating Value	P4	Trapezoidal	[−436 −46.3 1770 2390]
	P3	Trapezoidal	[2390 2390 3580 4165]
	P2	Triangular	[3580 4165 4750]
	P1	Trapezoidal	[4165 4750 10,000 10,000]
Chlorine content	C1	Trapezoidal	[−1 −0.5265 0.5 1]
	C2	Triangular	[0.5 1 1.5]
	C3	Trapezoidal	[1 1.5 3 3]
	C4	Triangular	[3 3.5 5.167]
Mercury content	H1	Trapezoidal	[−0.2 −0.1032 0.1 0.175]
	H2	Triangular	[0.1 0.175 0.25]
	H3	Trapezoidal	[0.175 0.25 0.5 0.5]
	H4	Triangular	[0.5 0.7 1]

presents the definition of the degree of membership for the input variables.

The system’s output variable was named “RDF Classification”, and a real number was generated within the interval [0, 1]. Trapezoidal membership functions were used, and the degrees of membership for the output variable are shown in

Table 6. Definition of the degree of membership of the output variable.

Fuzzy Set	Type of Function	Limiting Factor
Not an RDF	Trapezoidal	[−0.375 −0.1 0.1 0.2]
Terrible	Trapezoidal	[0.2 0.2 0.3 0.4]
Bad	Trapezoidal	[0.3 0.4 0.5 0.6]
Regular	Trapezoidal	[0.5 0.6 0.7 0.8]
Good	Trapezoidal	[0.7 0.8 0.9 1]
Excellent	Trapezoidal	[0.9 1 1 10]

and Figure 4.

Since each of the three input variables has 4 classes, a total of sixty-four (4 × 4 × 4) combinations were made between the fuzzy sets. To construct the system’s rule base, which will be presented in Section 4 (Table 8), expert analysis was considered. This analysis was collected through a structured questionnaire to rate the RDF as “EXCELLENT”, “GOOD”, “REGULAR”, “BAD”, “TERRIBLE”, or “NOT AN RDF”, based on the interactions of the classes proposed for the RDFs (Table 4), adapted from [25].

The Mamdani inference method was used to calculate the numerical value of the output variable. For defuzzification, the center of mass (or center of gravity) method was applied.

The decision to adopt fuzzy logic was guided by several factors inherent to waste characterization challenges. Fuzzy systems allow the integration of qualitative expert knowledge through linguistic rules, which is particularly valuable in scenarios where data may be scarce, noisy, or imprecise—as is common in Municipal Solid Waste (MSW) analysis. Unlike machine learning models that require extensive training datasets, fuzzy models are easily interpretable and adaptable to changing parameters, enabling rapid adjustments without retraining. Moreover, fuzzy logic systems offer computational simplicity, making them suitable for localized or resource-limited contexts, such as small municipalities. As highlighted by [28], fuzzy logic mimics human reasoning more naturally than binary logic, which is advantageous when dealing with vague descriptors like “high heating value” or “low chlorine content”. These qualities make fuzzy logic especially well suited for supporting decision-making in RDF classification under uncertainty.

3.3. Characterization of Samples Collected in the Municipality of Tupã

To verify the effectiveness of the model, the classification model was applied to RDF samples from the municipality of Tupã, São Paulo, Brazil. Different percentages of peanut shells were incorporated into some of these samples to assess whether the inclusion of this type of material could genuinely improve the quality of the RDF.

3.3.1. Sample Collection and Preparation

The only material used in this study was the waste collected from Municipal Solid Waste (MSW) in the municipality of Tupã, São Paulo, Brazil. All recyclable fractions were removed during the collection process. At the transfer station, the material was stored in four containers. As soon as a container was full, it was transported to a landfill in another municipality located approximately 45 km away. On the collection day, only one of the containers contained waste, so it was decided to collect waste from five different points to ensure variability in waste removal.

The material was then weighed on a BAISEC-brand digital electronic scale, model with a 40 kg capacity, until a mass of 5 kg was reached. Any excess was collected and properly discarded.

On the same day, an agro-industry in the municipality provided 5 kg of peanut shells from its production process to be used as samples for this study.

To prepare the samples, 500 g of peanut shells were first ground using a MARCONI knife mill.

For the MSW, to establish a homogenization standard for the subsequent grinding of 500 g of sample, two consecutive quarterings were performed. The quartering process, as outlined by NBR 10.007 [29], involves dividing the sample into four equal parts. Two opposite parts are selected to represent the whole sample, while the remaining two aliquots are discarded. The quartering process can be repeated as many times as necessary until the desired mass for subsequent analyses is achieved.

Next, a magnet was used to remove any magnetic materials present in the waste samples, such as screws, to prevent damage to the equipment. The material was then crushed using a SEIBT knife mill, model MGHS6/230.

Subsequently, different percentages of waste were mixed with the crushed peanut shells, based on the proportions from the study conducted by [24], to create five different types of samples, each weighing 20 g. A MARTE analytical balance, model AY220, was used for this purpose.

Table 7. Sample composition.

Sample	Refuse Percentage	Peanut Shell Percentage	Refuse Mass	Peanut Shell Mass
A1	100%	0	20 g	0
A2	90%	10%	18 g	2 g
A3	75%	25%	15 g	5 g
A4	50%	50%	10 g	10 g
A5	0	100%	0	20 g

presents the different compositions of the five samples.

After preparing the samples, all in triplicate, the process of determining the heating value, mercury content, and chlorine content was initiated, as described below.

3.3.2. Determination of Lower Heating Value

To determine the lower heating value (LHV), it is first necessary to determine the moisture content of the material. For this purpose, an infrared moisture balance from SHIMADZU, model MOC63u, was used.

0.5 g of each of the five samples were placed in the equipment to determine the moisture content, and the results will be presented in a later section.

Next, a Parr calorimeter, model 6400, was used to determine the lower heating value of the samples. The moisture content of the sample, as obtained in the previous step, is entered into this device.

Due to the high moisture content of the sample with pure waste, it was necessary to use the sample in its granulated form, rather than the solid pellet form typically used in calorimeters. For this, 0.5 g of each sample was used, and analyses were performed in triplicate.

It is worth mentioning that for the sample composed solely of waste (A₁), the material did not ignite after three attempts. The ignition wire burned, but the compound, which was the subject of this analysis, did not catch fire, likely due to the high moisture content of the waste. Therefore, to obtain the LHV of A₁, an auxiliary drying process was employed using an IMARVIL oven for 4 h at 100 °C.

After the sample was removed from the oven, it was taken to the calorimeter, and the moisture value of the still-wet material was entered into the equipment to calculate the LHV of A₁.

3.3.3. Determination of Chlorine Content

The X-ray fluorescence (XRF) method was used to determine the chlorine content of the sample, utilizing Shimadzu equipment, model XRF700 (Singapore), which has scanning capabilities from Sodium (Na) to Scandium (Sc) and from Aluminum (Al) to Uranium (U). This equipment identifies the chemical composition of samples, i.e., the constituent elements. The analysis was performed in triplicate using 0.5 g of each sample.

After the samples were placed in the equipment, analyses were performed, and the results were recorded in the form of a table, which will be presented in a later section.

3.3.4. Determination of Mercury Content

The total mercury content in the sample was determined in triplicate using a methodology developed by [30], employing a PERKIN ELMER atomic absorption spectrometer, model SMS 100 (Singapore), designed to measure mercury content in solid samples [31].

4. Results

4.1. Fuzzy Modeling for RDF Classification

Table 8. Fuzzy rule base for classifying RDF.

Rule	Input Variables			Output Variable
	LHV	Chlorine Content	Mercury Content	RDF Classification
1	P1	C1	H1	EXCELLENT
2	P1	C1	H2	GOOD
3	P1	C1	H3	REGULAR
4	P1	C1	H4	Not an RDF
5	P1	C2	H1	GOOD
6	P1	C2	H2	GOOD
7	P1	C2	H3	REGULAR
8	P1	C2	H4	Not an RDF
9	P1	C3	H1	REGULAR
10	P1	C3	H2	REGULAR
11	P1	C3	H3	BAD
12	P1	C3	H4	Not an RDF
13	P1	C4	H1	Not an RDF
14	P1	C4	H2	Not an RDF
15	P1	C4	H3	Not an RDF
16	P1	C4	H4	Not an RDF
17	P2	C1	H1	GOOD
18	P2	C1	H2	GOOD
19	P2	C1	H3	REGULAR
20	P2	C1	H4	Not an RDF
21	P2	C2	H1	GOOD
22	P2	C2	H2	REGULAR
23	P2	C2	H3	BAD
24	P2	C2	H4	Not an RDF
25	P2	C3	H1	REGULAR
26	P2	C3	H2	REGULAR
27	P2	C3	H3	BAD
28	P2	C3	H4	Not an RDF
29	P2	C4	H1	Not an RDF
30	P2	C4	H2	Not an RDF
31	P2	C4	H3	Not an RDF
32	P2	C4	H4	Not an RDF
33	P3	C1	H1	REGULAR
34	P3	C1	H2	REGULAR
35	P3	C1	H3	BAD
36	P3	C1	H4	Not an RDF
37	P3	C2	H1	REGULAR
38	P3	C2	H2	BAD
39	P3	C2	H3	BAD
40	P3	C2	H4	Not an RDF
41	P3	C3	H1	BAD
42	P3	C3	H2	BAD
43	P3	C3	H3	VERY BAD
44	P3	C3	H4	Not an RDF
45	P3	C4	H1	Not an RDF
46	P3	C4	H2	Not an RDF
47	P3	C4	H3	Not an RDF
48	P3	C4	H4	Not an RDF
49	P4	C1	H1	Not an RDF
50	P4	C1	H2	Not an RDF
51	P4	C1	H3	Not an RDF
52	P4	C1	H4	Not an RDF
53	P4	C2	H1	Not an RDF
54	P4	C2	H2	Not an RDF
55	P4	C2	H3	Not an RDF
56	P4	C2	H4	Not an RDF
57	P4	C3	H1	Not an RDF
58	P4	C3	H2	Not an RDF
59	P4	C3	H3	Not an RDF
60	P4	C3	H4	Not an RDF
61	P4	C4	H1	Not an RDF
62	P4	C4	H2	Not an RDF
63	P4	C4	H3	Not an RDF
64	P4	C4	H4	Not an RDF

P—Lower Heating Value; C—Chlorine Content; H—Mercury Content.

presents the rules adopted in the modeling process based on the experts’ responses.

Figure 5, Figure 6 and Figure 7 represent the 3D surface formed as a solution to the fuzzy modeling by Mamdani Inference, in which the relationships between the input variables—LHV, chlorine content, and mercury content, respectively—and the output variable, RDF Classification, are presented.

4.2. Samples Collected in the Municipality of Tupã

4.2.1. Heating Value

Table 9. Moisture content and LHV of RDF samples from the municipality of Tupã.

Samples	Moisture (%)	LHV (cal·g−1)
A1 *	57.38	5403.69
A2	43.17	1962.35
A3	32.03	2736.45
A4	30.54	3287.14
A5	10.34	3962.26

* After passing through the furnace.

shows the moisture content and lower heating value (LHV) for 0.5 g samples of waste and peanut shells collected in the municipality of Tupã.

Next, the results of the chlorine and mercury content analyses will be presented.

4.2.2. Chlorine Content

Table 10. Chemical composition of RDF samples from the municipality of Tupã (%) found through XRF analysis.

Chemical Compound	A1 (%)	A2 (%)	A3 (%)	A4 (%)	A5 (%)
CaO	49.214	37.830	37.583	32.391	23.338
Cl	*	29.604	*	*	*
K2O	19.063	17.145	21.155	25.515	24.494
SiO2	15.703	5.768	19.766	18.091	28.276
Fe2O3	8.328	3.660	10.877	10.778	9.766
SO3	4.562	1.598	6.287	7.548	8.111
TiO2	1.320	3.516	2.169	2.494	2.535
MnO	0.432	0.236	0.682	0.626	0.641
CuO	0.375	0.168	0.759	0.487	0.523
ZnO	0.346	0.138	0.405	0.235	0.180
NiO	0.337	*	*	*	*
SrO	0.116	0.047	0.242	0.169	0.083
Br	0.087	0.108	*	*	*
P2O5	*	0.153	*	1.251	1.203
Cr2O3	*	*	*	0.246	0.767
V2O5	0.046	*	*	0.087	*
Rb2O	0.042	0.014	*	0.046	*
ZrO2	0.029	0.016	0.075	0.037	*
OsO4	*	*	*	*	0.082

* Below the equipment’s limit of quantification [32].

shows the chemical composition of the 0.5 g samples of waste and peanut shells collected in Tupã, as determined through X-ray Fluorescence (XRF) analysis.

4.2.3. Mercury Content

As a result, a total mercury content of 31.7409 × 10⁻⁶ mg·kg⁻¹ was found for sample 1, confirming the initial assumption that the total mercury content would be much lower than 0.1 mg·kg⁻¹. Therefore, all samples were classified as H₁.

4.3. Scenario Involving Samples Collected from the Municipality of Tupã

Based on the analyses of LHV, chlorine content, and mercury content, the results for the five samples collected from the municipality of Tupã are summarized in

Table 11. Analysis of samples collected at Tupã.

Samples	Mass (g)	LHV (kcal·kg−1)	Chlorine (%)	Mercury (mg·kg−1)	Moisture (%)
A1	0.5142	5403.69 *	**	31.7409 × 10−6	57.38
A2	0.5309	1962.35	29.60	**	43.17
A3	0.5254	2736.45	**	**	32.03
A4	0.5265	3287.14	**	**	30.54
A5	0.5306	3962.26	**	**	10.34

A₁—100% refuse; A₂—90% refuse; A₃—75% refuse; A₄—50% refuse; A₅—0% refuse; * After going through the furnace; ** Below the equipment’s limit of quantification [31,32].

.

It is worth noting that the high chlorine content observed in Sample A₂ may be related to the significant heterogeneity of municipal solid waste, which can exhibit considerable variations in its composition. This factor may explain the anomalous value, representing a possible cause for this outlier.

From Table 11, the scenarios were simulated based on the fuzzy model, as presented in Section 5.

Table 12. Summary of classifications performed by fuzzy modeling.

Samples	Rule Base	Degree of Membership	Classification
A1 *	1	0.97	EXCELLENT
A2	61	0.0899	NOT AN RDF
A3	33	0.65	REGULAR
A4	33	0.65	REGULAR
A5	17	0.778	GOOD
	33	0.222	REGULAR

* After going through the furnace.

summarizes the classification results obtained from fuzzy modeling for the samples from Tupã, SP.

5. Discussion

The 3D figures (Figure 5, Figure 6 and Figure 7) illustrate the interaction between two input variables in RDF classification. Figure 5 demonstrates that the higher the LHV (Lower Heating Value) and the lower the chlorine content, the better the RDF classification. Similarly, Figure 6 highlights that the higher the LHV and the lower the mercury content, the better the RDF classification. It is essential to note that for LHV levels below 2390 kcal·kg⁻¹, regardless of chlorine and mercury contents (as shown in both figures), the compound does not receive an RDF classification, as it is not considered technically viable for combustion. This finding underscores the unfeasibility of the fuel combustion process for such materials. In cases where a compound is deemed unfeasible, as outlined in Table 1 proposed by [23], an auxiliary fuel is needed to increase the LHV and transform the compound into an RDF. The data presented in these figures further support the classification suggested by [22], which was adapted from the American standard ASTM E856-83 and the ERFO website (see Table 3).

The 3D graph in Figure 7, however, shows results that are opposite to those in Figure 5 and Figure 6. According to NBR 16.849/2020 [25], the limits for the compound to be considered RDF are higher for chlorine (Cl > 3.0%) and mercury (Hg > 0.5 mg·kg⁻¹) content levels and lower for LHV (LHV < 2390 kcal·kg⁻¹). The data from these three surface graphs reveal clear and well-defined classification levels.

Table 9 shows that as the amount of peanut shells in the samples increased, their moisture content decreased and LHV increased, except for the LHV of A₁, which underwent a drying process to reduce moisture so that the RDF could be burned, as previously reported.

The data in Table 9 corroborate [24], suggesting that the mixture of sugarcane straw and eucalyptus bark with RDF improves fuel quality, particularly in terms of heating value and moisture content. Therefore, it can be inferred that combining RDF with drier biomass, i.e., materials with lower moisture content, results in a compound with a higher heating value, thereby enhancing its energy viability as a fuel.

Table 10 shows that sample 2 was the only one with a chlorine content of 29.6%. This can be attributed to the highly heterogeneous composition of the samples, as demonstrated in Figure 1 (percentage of each component of the MSW in Brazil) and further confirmed by Figure 2 (percentage of recyclable materials found in the municipality of Tupã). The variability in the composition may have contributed to the higher chlorine content in sample 2 compared to the other samples.

Based on an LHV of 5403.69 kcal·kg⁻¹, chlorine content of 0%, and mercury content of 31.7409 × 10⁻⁶ mg·kg⁻¹, the system classifies the RDF with a degree of membership of 0.97. Analyzing the degree of membership of the output variable reveals that it holds the highest membership within the optimal fuzzy set, as it falls under Rule 1 (Table 8). It is important to note that this ideal classification was only achievable due to the prior drying of the material, which aligns with the data presented in Table 1 [23] (p. 28), emphasizing the necessity of pre-treatment to enhance the LHV when waste combustion lacks energy viability.

Based on an LHV of 1962.35 kcal·kg⁻¹, chlorine content of 29.6%, and mercury content of 0 mg·kg⁻¹, the system classifies the RDF with a degree of membership of 0.0899. The highly heterogeneous composition of the samples may have contributed to the higher chlorine content in sample 2, as certain materials present in this sample were not found in the others, thus increasing the chlorine content.

Furthermore, it should be noted that this point has a higher degree of membership within the fuzzy set “Not an RDF”, as it corresponds to Rule 61 (Table 8). This result aligns with the classification in Table 4, where the LHV is classified as P4 and the chlorine content as C4, meaning that the sample cannot be considered RDF.

If the mercury content value were to reach 31.7409 × 10⁻⁶ mg·kg⁻¹, as in A₁, the RDF classification would remain unchanged (“Not an RDF”), since the rule base would not be affected (Rule 61, as shown in Table 8). The only difference would be in the degree of membership, which would decrease to 0.0814.

Another simulation was performed with an LHV of 2736.45 kcal·kg⁻¹, chlorine content of 0%, and mercury content of 0 mg·kg⁻¹, resulting in an improved RDF classification. This led to a “Regular” response based on the rule base shown in Table 8, with a degree of membership of 0.65).

Although chlorine and mercury contents are within their optimal ranges, C1 and H1, respectively, the LHV falls into the worst range, P3, as shown in Table 4 (adapted from [25]). As a result, the LHV negatively impacts the overall quality and, consequently, the RDF classification.

If the mercury content value were 31.7409 × 10⁻⁶ mg·kg⁻¹, as in A₁, the RDF classification would remain the same (“Regular”), since the rule base would not change (Rule 33) as presented in Table 8, and neither would the degree of membership (0.65).

In sample 4, the LHV was 3287.14 kcal·kg⁻¹, the chlorine content was 0%, and the mercury content was 0 mg·kg⁻¹. The scenario for sample 4 was very similar to that for sample 3, as the degree of membership was the same (0.65), resulting in the system response being identical: “Regular” (Table 8). Therefore, increasing the LHV from 2736.45 kcal·kg⁻¹ to 3287.14 kcal·kg⁻¹ while maintaining chlorine and mercury content at zero did not affect the RDF classification.

A simulation was also conducted to determine the effect of a mercury content of 31.7409 × 10⁻⁶ mg·kg⁻¹ in sample 4. In this scenario, with the mercury content at 31.7409 × 10⁻⁶ mg·kg⁻¹, the rule base, as shown in Table 8, would remain unchanged (Rule 33), and the degree of membership (0.65) would also stay the same. Therefore, the RDF classification would remain unchanged as “Regular”.

In the scenario of sample 5, the system simulation activated two rule bases—Rule 17 and Rule 33 (see Table 8)—resulting in two possible classifications: “Good” and “Regular”. This indicates that the sample falls within an intermediate condition. The final classification was determined to be 0.778. Upon analyzing the degrees of membership for the output variable, it was observed that the sample had a higher degree of membership within the “Good” fuzzy set, thus leading to its classification as “Good”.

Ultimately, a simulation was conducted for sample 5 to assess the impact of a mercury content of 31.7409 × 10⁻⁶ mg·kg⁻¹. As observed when the mercury content is zero, this value falls within the same classification range. Consequently, the rule bases (Rules 17 and 33 from Table 8) remained unchanged, as did the degree of membership (0.778). Therefore, the RDF classification continued to be “Good”.

As shown in Table 12, only sample 1 achieved an “Excellent” RDF classification, as it was composed exclusively of refuse-derived fuel. However, as previously noted, the material had to be dried in order to determine its LHV. This supports the assertion made by [23] in Table 1, which states that when the LHV is insufficient, thermal pre-treatment is necessary.

Despite the success achieved with sample 1 after undergoing furnace drying at the laboratory scale, it is likely that, at the industrial level—whether in incineration plants, boilers, clinker kilns, or other facilities—the cost of pre-drying the material would render the process economically unfeasible. This is primarily because drying is among the most significant operational expenses in such industries. Nevertheless, further studies are needed to validate this assumption, particularly those exploring alternative methods such as solar drying.

Another alternative when there is no technical feasibility for burning the material is the addition of an auxiliary fuel to the compound [23]. In samples 2, 3, and 4, fractions of peanut shells were incorporated to evaluate the increase in LHV and, consequently, the potential improvement in RDF classification.

In sample 2, despite the addition of 10% peanut shells, there was no improvement in LHV classification compared to sample 1, as it remained in range P4 (LHV < 2390 kcal·kg⁻¹). Moreover, the chlorine content was notably high (29.6%), placing it in the worst classification range (C4). It is important to note that having even a single variable in range 4 is sufficient to classify the compound as “Not an RDF”.

In samples 3 (with 25% peanut shells) and 4 (with 50% peanut shells), the RDF classification remained “Regular”. This outcome is due to both chlorine and mercury contents staying within their optimal ranges—C1 and H1, respectively—making the LHV the deciding factor for classification. Although the proportion of peanut shells was doubled, the LHV in both cases remained within the P₃ range (2390 kcal·kg⁻¹ ≤ LHV < 3580 kcal·kg⁻¹). Therefore, in this context, increasing the amount of peanut shells did not significantly enhance the LHV or improve the overall fuel classification.

Furthermore, it was observed that sample 5, composed exclusively of peanut shells, showed a significant increase in LHV (3962.26 kcal·kg⁻¹), placing it within the P₂ range (3580 kcal·kg⁻¹ ≤ LHV < 4750 kcal·kg⁻¹). This led to an improved RDF classification of “Good”, as both chlorine and mercury contents remained within optimal levels (C₁ and H₁, respectively). However, the exclusive use of peanut shells raises concerns regarding volume and material availability. While municipal solid waste (MSW) is consistently generated year-round, peanut production is seasonal—occurring primarily during the rainy season and, to a lesser extent, in the dry season, which often requires irrigation. Additionally, climate change and an increasing frequency of extreme weather events (such as heavy rainfall, droughts, wildfires, and floods) can significantly affect agricultural yields, making the reliable supply of peanut shells uncertain.

Based on these results, it can be concluded that, although chlorine and mercury contents remained within their optimal ranges (C₁ and H₁), the high moisture content—and consequently low LHV—of the waste led to an RDF classification of only “Regular”, even with the addition of peanut shells as an auxiliary fuel. This outcome suggests that raw RDF samples (i.e., without prior drying treatment) are not suitable for use in furnaces and boilers under the conditions examined in this study.

The scenario observed in Brazil closely mirrors the findings of this study. According to data presented in Figure 1 [21], more than half (51.4%) of the average particle size composition of municipal solid waste (MSW) in Brazil consists of the wet fraction. This high moisture content significantly impacts the energy viability of RDF production.

Regarding the use of RDF after drying the material, predicting its application is challenging without conducting a financial feasibility analysis, which was beyond the scope of this study. However, future research is recommended to focus on the economic feasibility of drying waste, both with and without the addition of biomass, for use in furnaces and boilers. This is particularly relevant, as positive results were observed in laboratory-scale experiments under similar conditions.

6. Conclusions

This study provided new insights into the use of fuzzy logic for classifying the quality of Refuse-Derived Fuel (RDF) composed of Municipal Solid Waste (MSW), with or without the addition of peanut shells. The proposed fuzzy model demonstrated strong potential for assisting in RDF classification based on key variables—Lower Heating Value (LHV), chlorine content, and mercury content—producing results that closely align with expert evaluations.

The classification outcomes confirmed that untreated MSW samples generally result in RDFs of low energy quality due to their high moisture content. Although the inclusion of peanut shells improved the LHV, most combinations remained classified as “Regular”, emphasizing the limited effectiveness of co-combustion alone when pretreatment (drying) is not applied.

One key contribution was the confirmation that drying significantly enhances RDF quality, as shown in Sample A1. However, industrial-scale drying may face cost and feasibility barriers. Thus, one potential pathway to address this limitation is the integration of residual steam from the combustion process itself to pre-dry MSW, providing a self-sustaining and energy-efficient solution. Future studies should explore this proposition in greater detail, including alternatives such as centrifugation, chemical conditioning, mechanical pressing, or hybrid drying systems.

Furthermore, the application of systemic and hybrid analytical methodologies could improve RDF classification precision. Suggested future directions include (a) Multi-Criteria Analysis (MCA), (b) Computational Modeling and Simulation, (c) Artificial Neural Networks (ANN), (d) System Dynamics, (e) Advanced Statistical Analysis and Machine Learning, and (f) Scenario Methodology. Additionally, fuzzy-based combinatorial approaches—such as fuzzy-morphological analysis or fuzzy-brainstorming—could be explored to enhance decision-making in RDF quality assessments.

Lastly, integrating RDF classification into circular economy strategies could yield broader environmental and economic benefits. Future investigations might evaluate the valorization of RDF combustion byproducts, identifying new applications and markets aligned with sustainable development goals.

In summary, this study contributes an adaptable, rule-based model for RDF classification, laying the groundwork for more robust and interdisciplinary decision-support systems in the field of waste-to-energy technologies.