Production line control system and method with functions of fault diagnosis and restoration scheme output

Abstract

The invention provides a production line control system and method with fault diagnosis and restoration scheme output, and belongs to the technical field of production line fault judgment, and the method comprises the following steps: S1, collecting vibration data, temperature field data and material flow data of all equipment in a production line; s2, establishing a fusion feature matrix; s3, outputting a diagnosis result of the equipment through the equipment prediction model, and obtaining whether the equipment has a fault or not; s4, querying the fault knowledge graph, and analyzing a fault report of the equipment; s5, querying each fault propagation path; the probability sum of each fault propagation path is calculated, and the fault propagation path with the maximum probability sum is regarded as the required fault propagation path; and S6, based on the Monte Carlo tree, generating a plurality of repair schemes of the required fault propagation path, taking the repair scheme corresponding to the minimum repair cost as the optimal repair scheme, and outputting the optimal repair scheme. The system and the method not only can analyze whether the equipment has a fault, but also can provide an optimal repair scheme according to a fault propagation path.

Description

Technical Field

[0001] The present invention belongs to the technical field of production line fault judgment, and specifically relates to a production line control system and method with the functions of diagnosing faults and outputting repair solutions. Background Art

[0002] Automated production lines are widely used in modern industrial production, aiming to improve production efficiency, reduce labor costs, and ensure the stability of product quality.

[0003] Currently, the production line can only rely on sensors to monitor the working data of the equipment on the production line, so as to issue a warning when the data is abnormal and timely remind relevant personnel to repair the equipment on the production line.

[0004] The disadvantage of this kind of production line is that when a fault occurs, it cannot quickly provide a solution to repair the fault, resulting in too long time consumed from identifying the fault to restarting the production line, which reduces the production efficiency of the production line. Therefore, it is necessary to design a production line control system and method with the functions of diagnosing faults and outputting repair solutions, which can timely provide a repair solution to solve the fault when the equipment on the production line fails, save downtime, and improve the efficiency of the production line. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a production line control system and method with the functions of diagnosing faults and outputting repair solutions. This system and method can not only analyze whether the equipment is faulty, but also give the optimal repair solution according to the fault propagation path.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A production line control method with the functions of diagnosing faults and outputting repair solutions, and the specific steps are as follows:

[0008] S1: Collect the vibration data, temperature field data, and material flow data of all equipment in the production line;

[0009] S2: Based on the attention mechanism, establish a fusion feature matrix for the vibration data, temperature field data, and material flow data of a certain equipment;

[0010] S3: Based on the CNN-LSTM hybrid neural network, establish an equipment prediction model; input the fusion feature matrix into the equipment prediction model, and the equipment prediction model outputs the diagnosis result of the equipment. According to the diagnosis result, it is known whether the equipment is faulty;

[0011] S4: When the equipment has a fault, query the established fault knowledge graph and analyze to obtain the fault report of the equipment;

[0012] S5: The dynamic Bayesian network reads the topological relationship of the faulty devices in step S3; according to the topological relationship between the faulty devices, each fault propagation path is queried and numbered; based on the dynamic Bayesian network, the sum of probabilities of each fault propagation path is calculated, and the sum of probabilities of each fault propagation path is compared. According to the comparison result, the fault propagation path with the largest sum of probabilities is selected, and the fault propagation path with the largest sum of probabilities is regarded as the required fault propagation path. Risk assessment is carried out on the required fault propagation path, and the evaluated risk value is compared with the set risk threshold. According to the comparison result, it is decided whether to output the required fault propagation path;

[0013] S6: Based on the Monte Carlo tree, multiple repair schemes for the required fault propagation path output in S5 and the fault report in S4 are generated. The repair costs of each repair scheme are compared, and the repair scheme corresponding to the minimum repair cost is selected. The repair scheme corresponding to the minimum repair cost is the optimal repair scheme, and the optimal repair scheme is output.

[0014] As a further improvement of the present invention, in step S2, the formula for the fusion feature matrix is:

[0015]

[0016] In formula (1), α i (t) represents the weight of the i-th sensor at time t; x i (t) represents the data collected by the i-th sensor.

[0017] As a further improvement of the present invention, in step S3, the formula for the device prediction model is:

[0018] Diagnosis result =CNN-LSTM(Multi-feature vector ,Model parameters ,Threshold) (2)

[0020] In formula (2), Multi-feature vector is the vector representation of the fusion feature matrix X f (t);

[0021] Model parameters is the model parameter;

[0022] Threshold is the preset safety threshold;

[0023] Diagnosis result is the diagnosis result of the device.

[0024] The update formula for the neural network weight parameters in the device prediction model is:

[0025]

[0026] In formula (5), y(t) represents the actual state of the device;

[0027] represents the predicted state of the device;

[0028] represents the loss function;

[0029] represents the gradient;

[0030] η is the learning rate;

[0031] θ(t) represents the neural network weight parameters at the current moment;

[0032] θ(t + 1) represents the neural network weight parameters at the next moment.

[0033] As a further improvement of the present invention, the device prediction model uses VAE to determine whether the fused feature matrix is incorrect;

[0034] The reconstruction error formula of VAE is as follows:

[0035]

[0036] In formula (6), e(t) represents the reconstruction error;

[0037] x i (t) represents the data collected by the i-th sensor of the device; represents the predicted value of the data collected by the i-th sensor of the device;

[0038] If e(t) is greater than the set error threshold R, the fused feature matrix is incorrect; otherwise, the fused feature matrix is correct; when the fused feature matrix is incorrect, recalculate the fused feature matrix and do not continue to judge whether the device has a fault.

[0039] As a further improvement of the present invention, in step S5, the risk assessment formula for the required fault propagation path is:

[0040] R 风险 = E T 停机 P 次生故障 C 质量 (3)

[0041] In formula (3), R 风险 represents the required fault propagation path risk value;

[0042] E[T 停机It represents the sum of the repair times of each device on the required fault propagation path;

[0043] P 次生故障 It represents the sum of the probabilities of each device on the required fault propagation path;

[0044] C 质量 It represents the sum of the quality impact coefficients of each device on the required fault propagation path;

[0045] If R 风险 is greater than or equal to the set risk threshold, then output the required fault propagation path;

[0046] If R 风险 is less than the set risk threshold, then do not output the required fault propagation path; repeat S1 - S5.

[0047] As a further improvement of the present invention, in step S6, the repair cost formula of the device on the required fault propagation path is:

[0048]

[0049] In formula (4), k represents the number of devices in the required fault propagation path; T z represents the time cost of the method adopted by the z-th device on the required fault propagation path; N z represents the energy consumption cost of the method adopted by the z-th device; M z represents the quality loss cost of the method adopted by the z-th device.

[0050] The present invention provides a production line control system with the functions of diagnosing faults and outputting repair solutions, including the following modules:

[0051] A data acquisition module, which is used to acquire the vibration data, temperature field data, and material flow data of the device;

[0052] A feature fusion module, which is used to generate a fusion feature matrix based on the vibration data, temperature field data, and material flow data of the device;

[0053] A fault diagnosis module, equipped with a neural network, receives the fusion feature matrix, and generates a diagnosis result of the device;

[0054] A fault path analysis module, which is used to receive the diagnosis result of the faulty device, and according to the topological relationship, analyze the fault propagation path of the faulty device; analyze the probability of the propagation path, and decide whether to output the required fault propagation path;

[0055] The solution generation module generates multiple repair solutions based on the Monte Carlo tree according to the required fault propagation path and in combination with the diagnosis results, calculates the repair cost of each repair solution, selects the repair solution corresponding to the minimum repair cost, records it as the optimal repair solution, and outputs the optimal repair solution.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] This solution not only predicts whether a device fails through a neural network, but also generates a device fault propagation path based on the topological relationship of the device after the device fails, calculates the probability of the fault propagation path through a dynamic Bayesian network, selects the fault propagation path with the maximum probability, indicating that the fault will propagate along this fault propagation path, and then performs a risk assessment on this fault propagation path. Depending on the comparison between the evaluated risk value and the set risk threshold, according to the comparison result, it is decided whether to repair this fault propagation path, and the fault urgency of relevant personnel can also be reminded according to the magnitude of the risk value; finally, in the case where the fault propagation path needs to be repaired, a solution with the minimum repair cost is given for the reference of relevant maintenance personnel in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flowchart of this control method;

[0059] Figure 2 is a logic block diagram of this control method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The present invention will be further described in detail below in conjunction with the drawings and the specific embodiments:

[0061] Embodiment 1: A production line control method with the functions of diagnosing faults and outputting repair solutions. The flowchart is as Figure 1 shown, and the logic block diagram is as Figure 2 shown, including the following steps:

[0062] S1: Collect vibration data, temperature field data, and material flow data of all devices in the production line.

[0063] Specifically, use an adaptive resonance frequency filter to collect the vibration data of the device, and use the empirical mode decomposition method to process the vibration data to eliminate the coupling interference.

[0064] Use an infrared thermal imager to obtain the temperature field data of the device.

[0065] Use an RFID reader to collect the material flow data of the device, that is, obtain the material processing progress on this device.

[0066] S2: Based on the attention mechanism, establish a fusion feature matrix for the vibration data, temperature field data, and material flow data of a certain device.

[0067] The devices for collecting the above device data, namely the above adaptive resonance frequency filter, infrared thermal imager, and RFID reader, are collectively referred to as sensors.

[0068] For a certain device on the production line, it is set that there are a total of N sensors for collecting the data of this device; at time t, the data collected by the i-th sensor is denoted as x i (t), i ∈ [1, N]; in this embodiment, it is set that there is only one sensor for collecting the vibration data, temperature data, and material data of a certain device, so N is 3.

[0069] The formula for the fusion feature matrix of this device is:

[0070]

[0071] In formula (1), α i (t) represents the weight of the i-th sensor at time t; it is calculated by the attention mechanism.

[0072] S3: Based on the CNN-LSTM hybrid neural network, establish a device prediction model; input the fusion feature matrix into the device prediction model, and the device prediction model outputs the diagnosis result of the device, and it is known whether the device is faulty according to the diagnosis result.

[0073] Specifically, the formula for the device prediction model is:

[0074] Diagnosis result =CNN-LSTM(Multi-feature vector ,Model parameters ,Threshold) (2)

[0076] In formula (2), CNN-LSTM represents a hybrid neural network model composed of a convolutional neural network (CNN) and a long short-term memory network (LSTM); before using the device prediction model, the device prediction model has been trained well with a large amount of data;

[0077] Multi-feature vector is the vector representation of the fusion feature matrix X f (t).

[0078] Diagnosis result is the diagnosis result of the device;

[0079] Model parameters are model parameters; including neural network weight parameters and bias parameters.

[0080] Threshold is a preset safety threshold.

[0081] After inputting the fused feature matrix X f (t) into the device prediction model, the device prediction model will output its predicted prediction matrix If the prediction matrix is greater than the set safety threshold Threshold, it indicates that the device has a fault.

[0082] Example M1: The vibration data, temperature field data, and material flow data of a certain device at time t are: 0.4, 60, and 2 respectively. After being predicted by the device prediction model, the predicted vibration data, temperature field data, and material flow data are: 0.5, 63, and 3 respectively; then compare 0.5, 63, and 3 with the corresponding thresholds respectively. If the thresholds are: 0.8, 62, and 5 respectively, it can be known from the comparison results that the temperature field data is higher than the corresponding threshold, indicating that the device has a fault.

[0083] The neural network weight parameters in the device prediction model are constantly changing to ensure the accuracy of the prediction of the device prediction model. The update formula for the neural network weight parameters in the device prediction model is:

[0084]

[0085] In formula (3), y(t) represents the actual state of the device;

[0086] represents the predicted state of the device;

[0087] represents the loss function;

[0088] represents the gradient;

[0089] η is the learning rate;

[0090] θ(t) represents the neural network weight parameters at the current moment;

[0091] θ(t + 1) represents the neural network weight parameters at the next moment.

[0092] It is necessary to distinguish whether the input fused feature matrix is correct to avoid predicting an incorrect prediction matrix due to an incorrect fused feature matrix. Determine whether the fused feature matrix is incorrect through a variational autoencoder (VAE).

[0093] The reconstruction error formula of VAE is as follows:

[0094]

[0095] In formula (4), e(t) represents the reconstruction error;

[0096] x i (t) represents the data collected by the i-th sensor of the device; represents the predicted value of the data collected by the i-th sensor of the device;

[0097] Set the error threshold R;

[0098] If equation (4) is greater than or equal to R, it indicates that the input fusion feature matrix X f (t) is incorrect;

[0099] If equation (4) is less than R, it indicates that the input fusion feature matrix X f (t) is error-free. When the fusion feature matrix is incorrect, recalculate the fusion feature matrix and do not continue to judge whether the device has a fault.

[0100] Example M2:

[0101] Suppose the vibration data, temperature field data, and material flow data collected by a certain device at time t are: 0.4, 40, and 2 respectively. After being predicted by the device prediction model, the predicted vibration data, temperature field data, and material flow data are: 0.5, 63, and 3 respectively; set the error threshold R to 5, and according to equation (4), it can be calculated that

[0102] This 11.5 is greater than the set error threshold R, which indicates that the temperature field data collected by the device at time t is incorrect, and the prediction result of the device prediction model will not be output, and the data collected at this moment will be discarded.

[0103] S4: When the device has a fault, query the established fault knowledge graph and analyze to obtain the fault report of the device.

[0104] The definition of the knowledge graph is: it integrates and correlates the data related to faults in the form of a graph, so that the data related to faults no longer exist independently.

[0105] Taking M1 as an example, in M1, if the temperature field data of the device is higher than the set threshold, then according to the knowledge graph, the data about "the temperature of the device is too high" will be integrated into the fault report.

[0106] The fault report of this device is: {Device number: 001; Fault reason: Temperature is too high; Repair method: Replace the device or stop for inspection; Repairman: Li}

[0107] S5: The dynamic Bayesian network reads the topological relationship of the faulty devices in step S3; according to the topological relationship between the faulty devices, each fault propagation path is queried and numbered; based on the dynamic Bayesian network, the sum of probabilities of each fault propagation path is calculated, the sum of probabilities of each fault propagation path is compared, and according to the comparison result, the fault propagation path with the largest sum of probabilities is selected. The fault propagation path with the largest sum of probabilities is regarded as the required fault propagation path, and the required fault propagation path is subjected to risk assessment. The evaluated risk value is compared with the set risk threshold, and according to the comparison result, it is decided whether to output the required fault propagation path.

[0108] Example M3:

[0109] Suppose device A fails. Querying the topological relationship shows that device B and device C are respectively connected to device A and are secondary devices of device A; at the same time, device D and device E are respectively connected to device B and device C and are secondary devices of device B and device C respectively.

[0110] From the topological relationship, it can be seen that there are two fault propagation paths:

[0111] The first one: device A - device B - device D.

[0112] The second one: device A - device C - device E.

[0113] Set the prior probability of device A failing as P A , based on the dynamic Bayesian network, according to the probability P A of device A, the probability P B of device B failing is calculated, and then according to the probability P B , the probability P D of device D failing is calculated.

[0114] Similarly, set the prior probability of device A failing as P A , based on the dynamic Bayesian network, according to the probability P A of device A, the probability P C of device C failing is calculated, and then according to the probability P C , the probability P E of device E failing is calculated.

[0115] As above, the sum of probabilities of the two fault propagation paths are respectively:

[0116] The first one: P A + P B + P D ;

[0117] The second one: P A + P C + P E;

[0118] Compare the sum of probabilities of the first and the second one. If the sum of probabilities of the first one is greater than that of the second one, output the fault propagation path of the first one as the required fault propagation path; if the sum of probabilities of the second one is greater than that of the first one, output the fault propagation path of the second one as the required fault propagation path.

[0119] Conduct a risk assessment on the required fault propagation path. According to the assessment results, decide whether to output the required fault propagation path;

[0120] The risk assessment formula for the required fault propagation path is:

[0121] R 风险 = E T 停机 P 次生故障 C 质量 (5)

[0122] In formula (5), R 风险 represents the risk value of the required fault propagation path;

[0123] E[T 停机 represents the sum of the downtime repair times of each device on the required fault propagation path;

[0124] P 次生故障 represents the sum of probabilities of each device on the required fault propagation path;

[0125] C 质量 represents the sum of the quality impact coefficients of each device on the required fault propagation path.

[0126] If R 风险 is greater than or equal to the set risk threshold, output the required fault propagation path;

[0127] If R 风险 is less than the set risk threshold, do not output the required fault propagation path; repeat S1 - S5.

[0128] S6: Based on the Monte Carlo tree, generate multiple repair schemes for the required fault propagation path output in S5 and the fault report in S4. Compare the repair costs of each repair scheme, select the repair scheme corresponding to the minimum repair cost, and the repair scheme corresponding to the minimum repair cost is the optimal repair scheme, then output the optimal repair scheme.

[0129] Example M4:

[0130] Assume that the first fault propagation path is the required fault propagation path and is output. For devices A, B, and D, there are two solutions each, denoted as Solution One and Solution Two.

[0131] Then, the generated repair plan is shown in Table 1.

[0132] Table 1 Repair Plans for Equipment A, Equipment B, and Equipment C

[0133] Repair solution Device A adopts the solution Device B adopts the solution Device D adopts the solution Solution 1 Method 1 Method 1 Method 1 Solution 2 Method 1 Method 1 Method 2 Solution 3 Method 1 Method 2 Method 1 Solution 4 Method 1 Method 2 Method 2 Solution 5 Method 2 Method 1 Method 1 Solution 6 Method 2 Method 1 Method 2 Solution 7 Method 2 Method 2 Method 1 Solution 8 Method 2 Method 2 Method 2

[0134] The repair cost V = time cost T + energy consumption cost N + quality loss cost M.

[0135] Assume that the time costs of adopting Method 1 and Method 2 are respectively: L 1 and L 2 .

[0136] The energy consumption costs of adopting Method 1 and Method 2 are respectively: A 1 and A 2 ;

[0137] The quality loss costs of adopting Method 1 and Method 2 are respectively: F 1 and F 2 ;

[0138] The repair costs of each plan are:

[0139]

[0140] In Equation (6), T z represents the time cost of the method adopted by the z-th equipment in the required fault propagation path; k represents the number of equipment in the required fault propagation path;

[0141] N z represents the energy consumption cost of the method adopted by the z-th equipment; M z represents the quality loss cost of the method adopted by the z-th equipment.

[0142] Then: The repair cost of Plan 1; V1 = P1 * (L 1 + L 1 + L 1 ) + P2 * (A 1 + A 1 + A 1 ) + P3 * (F 1 + F 1 + F 1 ). In the formula, P1, P2, and P3 are all set coefficients, and the coefficients are constants, which are adjusted according to the actual situation.

[0143] The repair cost of Plan 2 V2 = P1 * (L 1 + L 1 + L 2 ) + P2 * (A 1 + A 1 + A 2 ) + P3 * (F1 +F 1 +F 2 );

[0144] And so on...

[0145] Select the minimum repair cost. The formula is as follows:

[0146] Q = min{V1, V2, V3, V4, V5, V6, V7, V8};

[0147] The solution corresponding to the minimum repair cost is the optimal repair solution.

[0148] Embodiment 2: A production line control system with the functions of diagnosing faults and outputting repair solutions, including the following modules:

[0149] A data acquisition module for acquiring vibration data, temperature field data, and material flow data of the equipment;

[0150] A feature fusion module for generating a fusion feature matrix based on the vibration data, temperature field data, and material flow data of the equipment;

[0151] A fault diagnosis module equipped with a neural network, receiving the fusion feature matrix, and generating a diagnosis result of the equipment;

[0152] A fault path analysis module for receiving the diagnosis result of the faulty equipment, analyzing the fault propagation path of the faulty equipment according to the topological relationship; analyzing the probability of the propagation path, and deciding whether to output the required fault propagation path;

[0153] A solution generation module, based on the required fault propagation path, combined with the diagnosis result, generating multiple repair solutions based on the Monte Carlo tree, calculating the repair cost of each repair solution, selecting the repair solution corresponding to the minimum repair cost, recording it as the optimal repair solution, and outputting the optimal repair solution.

[0154] The above is only a preferred embodiment of the present invention, and it is not any other form of limitation to the present invention. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A production line control method capable of diagnosing faults and outputting repair solutions, the specific steps are as follows, characterized in that: S1: Collect vibration data, temperature field data and material flow data of all equipment in the production line; S2: Based on the attention mechanism, the vibration data, temperature field data and material flow data of a certain equipment are used to establish a fusion feature matrix; S3: Based on the CNN-LSTM hybrid neural network, an equipment prediction model is established; the fusion feature matrix is ​​input into the equipment prediction model, and the equipment prediction model outputs the equipment diagnosis result, and whether the equipment is faulty is determined according to the diagnosis result; S4: When a device fails, the established fault knowledge graph is queried to analyze and obtain a fault report of the device; S5: The dynamic Bayesian network reads the topological relationship of the faulty devices in step S3; according to the topological relationship between the faulty devices, each fault propagation path is queried and numbered; Based on the dynamic Bayesian network, the probability sum of each fault propagation path is calculated, and the probability sum of each fault propagation path is compared. According to the comparison result, the fault propagation path with the largest probability sum is selected, and the fault propagation path with the largest probability sum is regarded as the required fault propagation path. The required fault propagation path is evaluated for risk, and the evaluated risk value is compared with the set risk threshold. According to the comparison result, it is decided whether to output the required fault propagation path; S6: Generate multiple repair plans for the required fault propagation path based on the Monte Carlo tree output of the required fault propagation path in S5 and the fault report in S4, compare the repair costs of each repair plan, and select the repair plan corresponding to the minimum repair cost. The repair plan corresponding to the minimum repair cost is the optimal repair plan, and the optimal repair plan is output.

2. The production line control system and method capable of diagnosing faults and outputting repair solutions according to claim 1, characterized in that: In step S2, the formula for the fusion feature matrix is: In formula (1), α i (t) represents the weight of the i-th sensor at time t; x i (t) represents the data collected by the i-th sensor.

3. The production line control method capable of diagnosing faults and outputting repair solutions according to claim 1, characterized in that: In step S3, the equipment prediction model formula is: Diagnosis result =CNN-LSTM(Multi-feature vector ,Model parameters ,Threshold) (2) In formula (2), Multi-feature vector is the fusion feature matrix X f Vector representation of (t); Model parameters are model parameters; Threshold is the preset safety threshold; Diagnosis result The diagnostic result of the device.

4. The production line control method capable of diagnosing faults and outputting repair solutions according to claim 3, characterized in that: The neural network weight parameter update formula in the device prediction model is: In formula (5), y(t) represents the actual state of the device; Indicates the predicted state of the device; represents the loss function; represents the gradient; η is the learning rate; θ(t) represents the neural network weight parameter at the current moment; θ(t+1) represents the neural network weight parameter at the next moment.

5. The production line control method capable of diagnosing faults and outputting repair solutions according to claim 3, characterized in that: The equipment prediction model uses VAE to determine whether the fusion feature matrix is ​​wrong; The reconstruction error formula of VAE is as follows: In formula (6), e(t) represents the reconstruction error; x i (t) represents the data collected by the i-th sensor of the device; Represents the predicted value of the data collected by the i-th sensor of the device; If e(t) is greater than the set error threshold R, the fused feature matrix is ​​wrong; Otherwise, the fused feature matrix has no errors; When the fused feature matrix is ​​wrong, the fused feature matrix is ​​recalculated and the determination of whether the device is faulty is not continued.

6. The production line control method capable of diagnosing faults and outputting repair solutions according to claim 1, characterized in that: In step S5, the risk assessment formula for the required fault propagation path is: R 风险 =E T 停机 P 次生故障 C 质量 (3) In formula (3), R 风险 represents the required fault propagation path risk value; E[T 停机 ] represents the time required for each device on the fault propagation path to be shut down for repair; P 次生故障 represents the probability and of each device on the required fault propagation path; C 质量 Represents the sum of the quality influence coefficients of each device on the required fault propagation path; If R 风险 If it is greater than or equal to the set risk threshold, the required fault propagation path is output; If R 风险 If the risk is less than the set risk threshold, the required fault propagation path is not output; and S1-S5 are repeated.

7. The production line control method capable of diagnosing faults and outputting repair solutions according to claim 1, characterized in that: In step S6, the required repair cost formula for the equipment on the fault propagation path is: In formula (4), k represents the number of devices in the required fault propagation path; T z N represents the time cost of the method adopted by the zth device on the required fault propagation path; z represents the energy consumption cost of the method adopted by the zth device; M z Represents the quality loss cost of the method adopted by the zth device.

8. A production line control system having the function of diagnosing faults and outputting repair solutions as claimed in any one of claims 1 to 7, characterized in that: Includes the following modules: Data acquisition module, used to collect vibration data, temperature field data and material flow data of equipment; Feature fusion module, used to generate a fusion feature matrix based on the vibration data, temperature field data and material flow data of the equipment; The fault diagnosis module is equipped with a neural network, receives the fused feature matrix, and generates the diagnosis results of the equipment; The fault path analysis module is used to receive the diagnosis results of the faulty device and analyze the fault propagation path of the faulty device according to the topological relationship; analyze the probability of the propagation path and decide whether to output the required fault propagation path; The solution generation module generates multiple repair solutions based on the Monte Carlo tree according to the required fault propagation path and the diagnosis results, calculates the repair cost of each repair solution, selects the repair solution corresponding to the minimum repair cost, records it as the optimal repair solution, and outputs the optimal repair solution.

Images