Fuzzy Logic-Based Fault Detection in Industrial Production Systems: A Case Study

The technological development of production technologies has led to the emergence of highly complex and dynamic industrial systems. Any industrial system can go directly from a normal operating mode to a defective mode in correspondence to a change or a stop in the industrial system ability to perform the production function. To avoid this problem, a systematic maintenance strategy can be followed through designing the system support, according to a certain pre-determined time period. Since this strategy requires a complete or a partial termination of the production system, it has a practical drawback, which in turn leads to a decrease in the productivity of the industrial system [1], [2]. To achieve this objective, it is necessary to find intelligent predictive technology that helps make decisions related to maintenance without the need to stop the production process [3], [4]. One of these techniques is the fuzzy logic technique. The intrinsic advantages of logic flow, according to the modeling approach, can be summarized in three points: reducing development cost [5], implementation and maintenance. The fact that in most organizations 60 to 80% of the physical activity provides maintenance makes the last feature the most important [6]. In general, system experts compare decision aid systems [7], [8], as well as traditional approaches [9], [10] and models based on flow logic or flow models [11], which have different properties:

• A strong reasoning engine.

• It is possible to abandon a prototype with significant delay [11].

• Adapting and auto regulator.

• Systems are designed to extend support in the face of contradictions [12] and the presence of ambiguous or imprecise information [13].

• The models are used for non-linear problems and their stability is enhanced in such contexts [14].

This study aims at developing the fuzzy logic technique in the field of diagnosis [15], [16], specifically detecting real industrial faults in compliance and dynamism and how to deal with them. For this purpose, the Omar Ben Omar Industrial Foundation, Guelma, Algeria, was proposed in this field of application, in particular, and circulated to institutions. In the industrial sector that practices the same activity, this introduction and case study can be summarized in the following dimensions; in order to achieve this objective, special attention is paid to the short pasta manufacturing process and the production line equipment (Line K) is analyzed, due to its specificity, which allows us to classify it, according to its criticality. Then, the functional or structural failure of the cash machine (press) is analyzed. Finally, fuzzy logic technology is developed. The study ends with a comprehensive conclusion summarizing the carried-out work as well as the alleged future work in this field.

As a part of the case study, the authors of this study are interested in the K production line for 4000 kg/h short pasta from the AMOR BENAMOR mill company in the wilaya of Guelma. The K line has an immediate availability and a fairly well-known stock level, due to an increase in consumption by Algerian customers. The current production capacity of the company is shown in Table 1.

It is made up of 8 pieces of equipment, as shown in Figure 1. The coding of the short pasta pattern K is shown in Table 2.

In order to quantify the equipment criticality in the production process of line K, an ABC analysis method has been used through taking the number of breakdowns for each machine as a criterion with four years of experience feedback [17], as shown in Table 3. According to the above table, the following diagram (Figure 2) has been constructed

According to the ABC analysis on the K line equipment, the chosen press is classified in zone A, and has a high percentage of breakdowns estimated at 59.31%. Table 4 shows the ABC table of the press. Figure 3 shows the ABC diagram of the press. In addition, the production process of the K line is a shop-type flow. Therefore, the press is a bottleneck machine. The press fault tree is presented in Figure 4. ABC analysis has been applied to the most sensitive part of production line K, which is the press, in order to know precisely at which level to apply fuzzy logic. The figure illustrates the results of the ABC analysis on the press components.

According to the ABC analysis of the press equipment, the dough-cutting equipment has been chosen for both. It is classified in zone A, and has a high breakdown percentage, estimated at 41.61%. Besides, it is highly sensitive, and has a repetitive type of failure. Therefore, the dough cutter has been chosen as the studied system in Figure 5. Failure probabilities are shown in Table 5.

In this application, membership functions need to be defined for each fuzzy subset of the three variables in Figure 6 and Table 6. Figure 7 shows the final function after the aggregation of fuzzy rules using a cut method.

Figure 8 presents the relationship between the failure rates A4 and A3 used for the degradation of E1.2 attributed to the paste cutting system:

• It can be noted that when the failure rate of the couple (A4 and A3) is strictly less than $3 \times 10^{-4}$, the system is always in operation or completely available. As the failure rate reaches its maximum, the dynamic behavior of the system becomes more stable.

• When the failure rate of the couple (A4 and A3) is greater than $3 \times 10^{-4}$, the two components (A3 and A4) are degraded successively up to a failure rate of $4.5 \times 10^{-4}$ torque, and the system is deteriorating (failing).

• Finally, when the failure rate of the couple (A4 and A3) is greater than $4.5 \times 10^{-4}$, t, the system enters the faulty phase, i.e., component E1.2 does not work or it’s completely broken down.

The failure rate of component E1.2 is stochastically dependent on the torque of the components (A4 and A3). The transition from failure A3, which is the lack of oil, influences the movement towards a more degraded status. That is the lack of lubrication, which is a factor contributing to the lowering of the capacity of the system and the increase of degradation rate of the affected E1.2 component. Therefore, the prognostic agent of component A4, which is the gears, is responsible for updating the shape parameter over time, depending on the established preventive maintenance program of the gears [18].

Figure 9 presents the relationship between the B2 bearing failure rates and the A3 lack of oil used for the degradation of E1.2 attributed to the paste cutting system.

• It can be noted that when the failure rate of the couple (B2 and A3) is strictly less than $3 \times 10^{-4}$, the system is always in operation or fully available. When the failure rate is at its maximum, the dynamic behavior of the system becomes more stable.

• When the failure rates of the couple (B2 and A3) are greater than $3 \times 10^{-4}$, the two components (B2 and A3) are gradually degraded up to a failure rate of $4.2 \times 10^{-4}$ of the torque, and the system is degrading (failing). When the value exceeds $4.2 \times 10^{-4}$, the system is suddenly degraded until it reaches the maximum value.

• When the component torque failure rates (B2 and A3) reach the maximum value $5 \times 10^{-4}$, the system does not work or completely fails.

The failure rate of component E1.2 is stochastically dependent on the torque of the components (B2 and A3). The transition from failure A3 (lack of oil) influences the movement towards more degraded status, and affects the degradation rate of component E1.2. Therefore, component prognosis agent B2 is responsible for updating the shape parameter over time based on the established bearing component preventive maintenance program [19], [20].

Complex industrial processes present highly non-linear dynamics and have a large number of variables. Therefore, sometimes it is difficult to obtain an accurate mathematical model to detect abnormal situations. Fuzzy logic technology has proven its effectiveness in detecting and predicting malfunctions before industrial systems stop. On the other hand, precautions for detection are necessary. Their performance depends strongly on a good selection of parameters and mastery of the process to be implemented. The authors are particularly interested in presenting the fuzzy approach within Omar Bin Omar and applying this approach to the K production line. First, the criticality of the machines on the K production line has been analyzed. Then, a fault tree has been developed to determine the causal relationships between malfunctions, failures and errors related to the critical machine. The ABC method has been used to find the most sensitive component of the press, the dough cutter, to which fuzzy logic could be applied. Finally, a fuzzy approach was applied to dough pieces in order to better detect anomalies as well as potential predictors of downtime and recovery. The obtained results are very satisfactory. This study can be generalized to institutions that contain the same industrial activity. As for future related studies, this approach could be generalized to all workshops in the factory. The idea of applying fuzzy logic to separate regulators is perhaps complicated and difficult to set up, but the simulation remains convincing. For the Omer ben Omor company, it is more interesting to have a fuzzy system of control, supervision, diagnosis and continuous maintenance than a system based on the stabilization of a complicated industrial system, which makes it difficult to control these parameters.