Fault root cause determination method and device and electronic equipment

Abstract

The invention discloses a fault root cause determination method and device and electronic equipment, and relates to the field of fault analysis. The method comprises the steps that a first Bayesian network is constructed based on key parameters of a first machine, a second Bayesian network is constructed based on key parameters of a second machine, and the abnormal degree of the first machine is different from that of the second machine; based on the first marginal probability and the first conditional probability of each node in the first Bayesian network and the second marginal probability and the second conditional probability of each node in the second Bayesian network, determining the marginal probability variable quantity and the causal relationship variable quantity of each node; based on the marginal probability variable quantity and the causal relationship variable quantity of each node and a preset weighting coefficient, determining a root cause score of each node; and determining the key parameter corresponding to the node with the highest root cause score in all the nodes as a fault root cause parameter. According to the invention, the troubleshooting time can be shortened, the production efficiency of a production line is improved, and the accuracy of fault positioning can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault analysis, and in particular to a method, device and electronic device for determining a root cause of a fault. Background Art

[0002] Various anomalies usually occur during the silicon wafer wire cutting process, which may lead to substandard product quality or even equipment shutdown, thus affecting production efficiency. Therefore, it is necessary to promptly investigate and handle the anomalies in the production process. Some of the anomalies in the production process can be solved by adjusting the process parameters, while the other part is caused by potential failures of the equipment itself. Due to the complex structure of the wire cutting machine, there are a large number of process parameters and components, so the root cause of the equipment failure is usually difficult to accurately locate.

[0003] The current troubleshooting method mainly relies on the experience of engineers. Engineers will make a preliminary judgment on the possible cause of the fault based on the working status, historical data and abnormal phenomena of the equipment. However, due to the subjectivity of experience and the complexity of the equipment, a large number of troubleshooting or disassembly experiments are usually required to finally determine the problem, which is not only time-consuming and easy to affect the overall efficiency of the production line, but also prone to misjudgment or missed judgment, which in turn leads to repeated abnormalities. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus and device for determining the root cause of a fault, so as to shorten the troubleshooting time and improve the accuracy of fault location.

[0005] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions: In a first aspect, a method for determining a root cause of a fault is provided, comprising: A first Bayesian network is constructed based on key parameters of a first machine, and a second Bayesian network is constructed based on key parameters of a second machine, wherein the abnormality of the first machine meets a preset condition, and the abnormality of the second machine exceeds the preset condition; Determine an edge probability change and a causal relationship change of each node based on a first edge probability and a first conditional probability of each node in the first Bayesian network and a second edge probability and a second conditional probability of each node in the second Bayesian network; Determine the root cause score of each node based on the edge probability change and the causal relationship change of each node and a preset weighting coefficient; The key parameter corresponding to the node with the highest root cause score among all the nodes is determined as the fault root cause parameter.

[0006] In some embodiments, the step of determining the edge probability change and the causal relationship change of each node includes: Determining an amount of change in the edge probability of each node based on a first edge probability of each node in the first Bayesian network and a second edge probability of a corresponding node in the second Bayesian network; Determine a change in the conditional probability of each node relative to the parent node based on a first conditional probability of each node in the first Bayesian network relative to any parent node and a second conditional probability of the corresponding node in the second Bayesian network relative to the corresponding parent node; The sum of the conditional probability changes of each node relative to all the parent nodes is determined as the causal relationship change of each node.

[0007] In some embodiments, the step of determining the causal relationship change amount of each of the nodes includes: The causal relationship change of each node is determined by the following formula:

[0008] Wherein, ΔC(X) is the change in the causal relationship of the node, X is the node, Y is the parent node, Pa(X) is the parent node set of node X, is the conditional probability change of the node X relative to the parent node Y.

[0009] In some embodiments, the preset weighting coefficient includes a first weighting coefficient and a second weighting coefficient; and the step of determining the root cause score of each of the nodes includes: The root cause score of each of the nodes is determined by the following formula:

[0010] Wherein, S(X) is the root cause score of node X, W1 is the first weight coefficient, W2 is the second weight coefficient, W1+W2=1; ΔP(X) is the edge probability change of node X, and ΔC(X) is the causal relationship change of node X.

[0011] In some embodiments, the first weight coefficient is 0.3, and the second weight coefficient is 0.7.

[0012] In some embodiments, constructing a first Bayesian network based on the key parameters of the first machine includes: Acquire historical data of each of the key parameters of the first machine, and divide the historical data into a training set and a test set; Based on the training set, using the maximum expectation method to determine the structure of the first Bayesian network and the direction of each edge; Determine a first conditional probability table for each node in the first Bayesian network using a maximum likelihood estimation method; The first Bayesian network is verified based on the test set, and the structure of the first Bayesian network and the direction of each edge are corrected according to the verification result to obtain the trained first Bayesian network.

[0013] In some embodiments, the first machine and the second machine are determined by the following steps: Determine the abnormality rate and abnormality rate variance of each operating machine in a preset period of time; Determine the operating machine whose abnormality rate is less than the first threshold and whose abnormality rate variance is less than the second threshold as the first machine; The operating machine whose abnormality rate is greater than or equal to the first threshold and / or whose abnormality rate variance is greater than or equal to the second threshold is determined as the second machine.

[0014] In some embodiments, the step of determining the key parameters comprises: Determine the equipment parameters of each critical control point of the operating machine, wherein the critical control point is used to characterize the equipment point that can affect the process; The key parameters are selected from the equipment parameters of each of the key control points.

[0015] In a second aspect, a device for determining a root cause of a fault is provided, comprising: A network construction module, configured to construct a first Bayesian network based on key parameters of a first machine, and to construct a second Bayesian network based on key parameters of a second machine, wherein the abnormality of the first machine satisfies a preset condition, and the abnormality of the second machine exceeds the preset condition; A network comparison module, for determining an edge probability change amount and a causal relationship change amount of each node based on a first edge probability and a first conditional probability of each node in the first Bayesian network, and a second edge probability and a second conditional probability of each node in the second Bayesian network; A score calculation module, used to determine the root cause score of each node based on the edge probability change and the causal relationship change of each node and a preset weighting coefficient; The root cause determination module is used to determine the key parameter corresponding to the node with the highest root cause score among all the nodes as the fault root cause parameter.

[0016] In a third aspect, an electronic device is provided, comprising a processor and a memory; a computer program is stored in the memory, and the processor is used to execute the computer program stored in the memory to implement the method for determining the root cause of a fault as described in any one of the first aspects.

[0017] One of the above technical solutions has the following advantages or beneficial effects: Compared with the prior art, a method for determining the root cause of a fault in the present application includes: constructing a first Bayesian network based on the key parameters of the first machine, and constructing a second Bayesian network based on the key parameters of the second machine, the abnormality of the first machine meets the preset conditions, and the abnormality of the second machine exceeds the preset conditions; based on the first edge probability and the first conditional probability of each node in the first Bayesian network, and the second edge probability and the second conditional probability of each node in the second Bayesian network, determine the edge probability change and the causal relationship change of each node; based on the edge probability change and the causal relationship change of each node, and the preset weighting coefficient, determine the root cause score of each node; determine the key parameters corresponding to the node with the highest root cause score among all nodes as the root cause parameters of the fault. The method for determining the root cause of a fault provided by the present application determines the root cause parameters of the fault by comparing the Bayesian networks of machines with different abnormality levels, which can not only shorten the troubleshooting time and improve the production efficiency of the production line, but also greatly improve the accuracy of fault location.

[0018] A fault root cause determination device of the present application determines the fault root cause parameters by comparing the Bayesian networks of machines with different abnormality levels, which can not only shorten the fault investigation time and improve the production efficiency of the production line, but also greatly improve the accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 This is a schematic diagram of the overall process of the method for determining the root cause of a fault according to an embodiment of the present application; Figure 2 Schematic diagram of the initial structure of the first Bayesian network in the embodiment of the present application; Figure 3 Schematic diagram of the initial structure of the second Bayesian network in the embodiment of the present application; Figure 4 A schematic diagram of the structure of a device for determining a root cause of a fault according to an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device according to an embodiment of the present application.

[0021] Reference numerals: 401 - network construction module; 402 - network comparison module; 403 - score calculation module; 404 - root cause determination module; 501 - memory; 502 - processor. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0023] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present application, "multiple" means two or more, and at least one means one, two or more, unless otherwise clearly and specifically defined.

[0024] Various anomalies usually occur during the wire cutting process of silicon wafers, such as broken blades, uneven cutting, surface defects, uneven cutting thickness, etc. Some anomalies can be solved by adjusting process parameters, such as adjusting cutting speed and tension, but other anomalies are caused by potential failures of the equipment itself. Due to the complex structure of the wire cutting machine, there are a large number of process parameters and components, which makes it difficult to accurately locate the source of the failure.

[0025] The current troubleshooting method mainly relies on the experience of engineers. Engineers will make a preliminary judgment on the possible causes of the failure based on the working status, historical data and abnormal phenomena of the equipment. However, although the experience of engineers can help locate the fault to a certain extent, due to the complex and changeable operating status of the equipment, it is often difficult to make accurate judgments based on experience, especially when faced with new abnormalities, which may lead to misjudgments or missed judgments. At the same time, due to the large number of variables involved in the wire cutting machine, when a failure occurs, a large number of troubleshooting or disassembly experiments are required to check the possible causes one by one. This process is not only time-consuming, but may also miss the golden time for production, further affecting the overall efficiency of the production line. In other words, the current troubleshooting method can often only find the surface causes of the problem, and it is difficult to find the real root cause of the failure.

[0026] In view of this, an embodiment of the present application provides a method for determining the root cause of a fault, which analyzes the complex relationship between the various components and parameters of the device and finds the real root cause of the fault based on the complex cause-and-effect chain within the device, thereby efficiently and accurately determining the root cause of the fault and avoiding the repeated occurrence of abnormalities, thereby solving at least part of the above-mentioned technical problems.

[0027] See also Figure 1 , Figure 1 The overall flow chart of the method for determining the root cause of a fault in an embodiment of the present application is as follows. The method for determining the root cause of a fault includes the following steps: Step 101: constructing a first Bayesian network based on key parameters of a first machine and constructing a second Bayesian network based on key parameters of a second machine, the abnormality level of the first machine meets a preset condition, and the abnormality level of the second machine exceeds the preset condition.

[0028] Specifically, key parameters refer to the parameters of key points that have a significant impact on the equipment process. It is understandable that for some equipment parameters, after adjustment, there will be no significant impact on the equipment process, so the equipment parameters are not key parameters.

[0029] In some examples, the step of determining the key parameters includes: Step 1: Determine the equipment parameters of each critical control point of the operating machine. The critical control points are used to characterize the equipment points that can affect the process.

[0030] Exemplarily, the operating machine may be a silicon wafer wire cutting device.

[0031] Step 2: Select key parameters from the equipment parameters of each key control point.

[0032] Specifically, the equipment parameters of each key control point that can have a greater impact on the process are determined as key parameters. There are many methods for selecting key parameters from the equipment parameters of each key control point. For example, the machine can be disassembled, and the key parameters can be selected based on the disassembly results and the mechanism of the machine. The key parameters can also be determined in advance according to actual needs. This embodiment of the application does not specifically limit this.

[0033] Exemplarily, key parameters may include cutting fluid flow (in 1 / min), cutting fluid temperature (in °C), left tension arm tension (in N), wire mesh speed (in m / s), left tension motor temperature (in °C), left spindle motor torque (in Nm), water-cooled motor inlet temperature (in °C), and feed speed (in mm / min). Among them, the cutting fluid can play a protective lubricating role when cutting the wire mesh, and the cutting fluid temperature can affect the protective lubricating ability of the cutting fluid. The cutting fluid flow rate can affect the lubrication and cooling effect of the cutting fluid. The tension of the left tension arm can control the tension or tightness of the cutting wire mesh. Too loose silicon wafers are prone to defects, and too tight wires are prone to breakage. The wire mesh speed can control the speed of the wire mesh going back and forth, and the speed will affect the cutting ability of the wire mesh. The left tension motor temperature can control the stability of the wire mesh tension. The left spindle motor torque can reflect the spindle driving force to ensure sufficient torque output during the cutting process. The water-cooled motor inlet temperature can control the temperature of the system to prevent overheating. The feed speed can affect the cutting depth and speed to ensure smooth and efficient cutting.

[0034] Through the above scheme, the key parameters selected have obvious influence on the process and may have a greater impact on the quality of the product. Therefore, including these key parameters in the monitoring scope can ensure the comprehensiveness and effectiveness of equipment operation monitoring.

[0035] In some embodiments, the degree of abnormality can be characterized by an abnormality rate and an abnormality rate variance. The first machine can represent a good machine, and the second machine can represent an out-of-tolerance machine. The first machine and the second machine can be determined by the following steps: Step 1: Determine the abnormality rate and abnormality rate variance of each operating machine in a preset period of time.

[0036] For example, the preset period can be set according to actual needs, for example, it can be set to one month. The abnormality rate is used to characterize the abnormal proportion of the running machine in the preset period. The abnormality rate can be determined by the following formula:

[0037] Among them, λ i is the abnormal rate of the i-th running machine in the preset period, i is an integer greater than or equal to 1 and less than or equal to I, I is the total number of running machines, n 异常 is the number of abnormal samples of the ith running machine in the preset period, n 总 is the total number of samples of the i-th running machine in the preset time period.

[0038] For example, taking the running machine as a wire cutting device, each time the running machine cuts a knife, it will record whether the knife is abnormal. The total number of cuts in the preset time period is the total number of samples of the running machine, and the abnormal knife times are the abnormal sample number of the running machine.

[0039] The abnormal rate variance is used to characterize the stability of the machine running within a preset period of time. The abnormal rate variance can be determined by the following formula:

[0040] in, is the variance of the abnormality rate of the ith running machine in the preset period, t is an integer greater than or equal to 1 and less than or equal to T, T is the total number of sub-periods into which the preset period is divided, λ t is the abnormal rate of the ith operating machine in the tth sub-period, is the average value of the abnormal rate of the i-th operating machine in each sub-period.

[0041] For example, the preset period is one month, and each sub-period is one week. In this way, the abnormality rate variance of one month can be determined according to the abnormality rate of the operating machine every week.

[0042] Step 2: Determine the operating machine whose abnormality rate is less than the first threshold and whose abnormality rate variance is less than the second threshold as the first machine.

[0043] Step three: determine the operating machine whose abnormality rate is greater than or equal to the first threshold and / or whose abnormality rate variance is greater than or equal to the second threshold as the second machine.

[0044] Exemplarily, the thresholds for different factories and different equipment may be different. The target value expected by the factory may be used as the threshold, or the threshold may be set in a clustering manner. This embodiment of the present application does not specifically limit this.

[0045] Through the above scheme, operating machines with low abnormality rate and small variance are classified as excellent machines, and operating machines with high abnormality rate and / or large variance are classified as out-of-tolerance machines. Therefore, operating machines can be classified according to the probability of abnormality occurrence and performance stability, thereby laying the foundation for subsequent inferences.

[0046] It is understandable that the key parameters of the first machine and the key parameters of the second machine are the same type of key parameters, such as cutting fluid flow, cutting fluid temperature, left tension arm tension, wire mesh speed, left tension motor temperature, left spindle motor torque, water cooling motor water inlet temperature, and feed speed. It is just that the relationship between these key parameters may change on machines with different abnormalities.

[0047] It can also be understood that the second machine can be used to characterize a machine operating in a fault state, and the first machine can be used to characterize a machine operating in a normal state.

[0048] See also Figure 2 , Figure 2It is a schematic diagram of the initial structure of the first Bayesian network in the embodiment of the present application. Wherein, a~l represent each node in the first Bayesian network, and each node represents a key parameter. The Bayesian network is a directed acyclic graph (DAG), which can model the cause-effect relationship between various variables. Exemplarily, node a can be the feed speed [mm / min], node b can be the temperature of the left active bearing box [℃], node c can be the cutting fluid flow [l / min], node d can be the temperature of the right active bearing box [℃], node e can be the temperature of the left tension motor [℃], node f can be the tension of the left tension arm [N], node g can be the speed of the left tension motor [rpm], node h can be the tension of the right tension arm [N], node i can be the cutting fluid temperature [℃], node j can be the temperature of the right tension motor [℃], node k can be the speed of the right tension motor [rpm], and node l can be the wire mesh speed [m / s].

[0049] In some embodiments, constructing the first Bayesian network may be achieved by the following steps: Step 1: Obtain historical data of key parameters of the first machine, and divide the historical data into a training set and a test set.

[0050] Specifically, the number of the first machines may be one or more, and the historical data of the key parameters of all the first machines are collected. The training set and the test set may be divided in a cross-validation manner.

[0051] In some examples, after obtaining historical data, the data can be first preprocessed to remove missing values ​​and outliers, and then the continuous variables can be discretized to meet the processing requirements of the Bayesian network. For example, the temperature can be divided into "high", "normal" and "low".

[0052] Step 2: Based on the training set, the maximum expectation method is used to determine the structure of the first Bayesian network and the direction of each edge.

[0053] Specifically, the maximum expectation method refers to the EM (Expectation-Maximization algorithm). The construction of the Bayesian network can be divided into two core parts: structure learning (determining the edges and their directions) and parameter estimation (determining the conditional probability). In the EM algorithm, the addition, deletion or retention of edges is achieved through repeated iterations of the E step and the M step, and the optimal network structure is finally determined.

[0054] Exemplarily, step 2 can be implemented by the following steps: The first step is to initialize the Bayesian network structure.

[0055] Specifically, each key parameter is a node in the Bayesian network. The initial structure can be assumed to be a graph without edges, all nodes are initially independent, or some of the initial edges can be defined based on existing domain knowledge.

[0056] The second step, the E step (expected step), is to calculate the conditional probability table of the current network.

[0057] Specifically, under the initial structure, the conditional probability table (CPT) of each node can be calculated, and the distribution of child nodes can be estimated according to the state of the parent node. Then the likelihood value of the current network is calculated, that is, the overall probability of the data under the current structure.

[0058] Exemplarily, the current network includes three nodes, namely, device temperature X1, vibration amplitude X2 and device operating status X3. By collecting data examples, the following Table 1 is obtained.

[0059] Table 1: Data example

[0060] First, calculate the edge probability of each node: P(X1=high)=3 / 5=0.6, P(X1=low)=2 / 5=0.4, P(X2=high)=2 / 5=0.4, P(X2=low)=3 / 5=0.6, P(X3=abnormal)=2 / 5=0.4, P(X3=normal)=3 / 5=0.6. Then calculate the probability of each sample. The probability of each sample is the product of the edge probability of the node. For example, the first sample P(X1=high, X2=high, X3=abnormal)=0.6*0.4*0.4=0.096. Finally, the product of all sample probabilities is determined as the total likelihood value of the current network. The total likelihood value is represented by the following formula:

[0061] Where L is the total likelihood value, n is an integer greater than or equal to 1 and less than or equal to N, N is the total number of samples, is the marginal probability of node X1 under the nth sample, is the edge probability of node X2 under the nth sample, is the marginal probability of node X3 under the nth sample.

[0062] The third step, M step (maximization step), is to adjust the edges.

[0063] Specifically, edge adjustment mainly includes adding, deleting and determining the direction of edges. The goal is to maximize the likelihood value of the current structure by adding and deleting edges.

[0064] Exemplarily, the steps of adding an edge mainly include: Choose two unbounded node pairs and assume that there is a potential causal relationship between the two nodes.

[0065] If there is prior knowledge, such as knowing that temperature affects the flow rate of cutting fluid, then an edge pointing from temperature to flow rate can be set. Determine the direction through the conditional independence test: If the conditional independence between node X and node Y cannot be explained by other variables, assume a direct edge from node X to node Y or node Y to node X. Use a scoring criterion (such as the Bayesian Information Criterion, BIC) to compare the two directions. Select the direction with a higher score and add this edge. After adding a new edge, recalculate the likelihood value of the network and update the conditional probability table (CPT) of each node. If the likelihood value increases and the increase value exceeds the set threshold, it means that this edge helps to improve the explanatory power of the network and retains the edge. If the likelihood value does not increase by more than the set threshold or the likelihood value decreases, the edge is not added.

[0066] Exemplarily, the steps of deleting an edge mainly include: Perform redundant edge detection, that is, try to temporarily delete existing edges and recalculate the likelihood of the network. If the likelihood of the network remains unchanged or increases slightly after deleting an edge, it means that the edge may be redundant and can be deleted. If deleting an edge causes the likelihood to decrease significantly, it means that the existence of this edge significantly increases the likelihood, and deleting this edge will reduce the explanatory power of the network, then keep the edge and finally traverse all edges to complete the likelihood evaluation.

[0067] Exemplarily, the steps of optimizing the direction of the edge mainly include: When there is an edge between two nodes, the final edge direction can be determined by comparing the likelihood values ​​in different directions. If the conditional probability table in a certain direction is more consistent with the data (i.e., the likelihood value is higher), the direction of the edge is set to this direction. Alternatively, domain knowledge can be used to determine the direction of the edge.

[0068] Step three, using the maximum likelihood estimation method to determine the first conditional probability table of each node in the first Bayesian network.

[0069] Specifically, the conditional probability can be calculated first, that is, the conditional probability of each node is calculated according to different combinations of parent nodes. For each parent node state, the probability distribution of each child node state is calculated. Then the maximum likelihood estimation (MLE) is performed, that is, the conditional probability of each node is maximized, so that the likelihood value of the entire network is maximized, and the final first conditional probability table is obtained.

[0070] Step 4: Verify the first Bayesian network based on the test set, and modify the structure of the first Bayesian network and the direction of each edge according to the verification result to obtain the trained first Bayesian network.

[0071] Specifically, verify the fitting effect of the first Bayesian network of the excellent machine on different data sets to ensure the stability of the structure. If the first Bayesian network performs poorly in the test set, the network can be optimized by modifying the network structure (re-adding or deleting edges, or changing the direction of edges). The focus can be on optimizing the connection and direction of the edges, and finally converge to the optimal network structure through repeated iterations.

[0072] Through the above method, the Bayesian network is introduced to deal with probability and uncertainty. Each key parameter is used as a node of the Bayesian network. By calculating its probability distribution, the first Bayesian network reflecting the excellent operation status of the machine is constructed. The first Bayesian network can not only predict the occurrence of events through conditional probability reasoning, but also reverse reasoning to determine the possible cause of a certain observation result. With the addition of new data, the network parameters can be dynamically updated through Bayesian theorem, so that the first Bayesian network can remain effective when the environment or conditions change.

[0073] See also Figure 3 , Figure 3 It is a schematic diagram of the initial structure of the second Bayesian network in the embodiment of the present application. Wherein, a~l represent each node in the second Bayesian network, and each node represents a key parameter. It can be seen that due to the different abnormalities of the first machine and the second machine, the causal relationship between each node has changed. For example, there is no direct causal relationship between node e and node f in the first Bayesian network, but there is a direct causal relationship in the second Bayesian network. Exemplarily, node a can be the feed speed [mm / min], node b can be the left active bearing box temperature [℃], node c can be the cutting fluid flow [l / min], node d can be the right active bearing box temperature [℃], node e can be the left tension motor temperature [℃], node f can be the left tension arm tension [N], node g can be the left tension motor speed [rpm], node h can be the right tension arm tension [N], node i can be the cutting fluid temperature [℃], node j can be the right tension motor temperature [℃], node k can be the right tension motor speed [rpm], and node l can be the wire mesh speed [m / s].

[0074] Accordingly, constructing the second Bayesian network can be achieved through the following steps: Step 1: Obtain historical data of each key parameter of the second machine, and divide the historical data into a training set and a test set.

[0075] Step 2: Based on the training set, the maximum expectation method is used to determine the structure of the second Bayesian network and the direction of each edge.

[0076] Step three, using the maximum likelihood estimation method to determine the second conditional probability table for each node in the second Bayesian network.

[0077] Step 4: Verify the second Bayesian network based on the test set, and modify the structure of the second Bayesian network and the direction of each edge according to the verification result to obtain the trained second Bayesian network.

[0078] It should be noted that the specific details of the above steps 1 to 4 can be found in the steps of constructing the first Bayesian network, which will not be repeated here.

[0079] It is understandable that by using Bayesian networks to characterize the operating status of the machine, it is possible to handle uncertainties that other acyclic graphs cannot directly handle, and it is more robust to noisy data. It can infer causal relationships through conditional probability distributions between nodes and quantify the possibility of each root cause. In addition, Bayesian networks can not only predict the occurrence of events, but also have strong reverse reasoning capabilities and can determine the possible causes of an observation. Moreover, with the input of new data, the network parameters can be dynamically updated through Bayesian theorem, so that the network can remain effective when the environment or conditions change, without the need for retraining, and dynamic adjustment is convenient. Therefore, Bayesian networks have a better combination effect.

[0080] Step 102: Based on the first marginal probability and the first conditional probability of each node in the first Bayesian network, and the second marginal probability and the second conditional probability of each node in the second Bayesian network, determine the marginal probability change and the causal relationship change of each node.

[0081] Specifically, marginal probability refers to the probability of a node occurring without considering other conditions. In a Bayesian network, the marginal probability of each node is calculated by accumulating all combinations of its parent node states. Conditional probability refers to the probability of a child node occurring given the parent node state. In a Bayesian network, edges represent causal relationships, and conditional probabilities are used to describe how child nodes depend on parent nodes.

[0082] In some embodiments, determining the edge probability change and causal relationship change of each node may be achieved by the following steps: Step 1: Determine the change in edge probability of each node based on the first edge probability of each node in the first Bayesian network and the second edge probability of the corresponding node in the second Bayesian network.

[0083] Specifically, the edge probability change of each node is determined by the following formula:

[0084] Among them, ΔP(X) is the change in the edge probability of node X, P2(X) is the second edge probability of node X in the second Bayesian network, which is obtained according to the parent node of node X in the second Bayesian network and the second conditional probability table; P1(X) is the first edge probability of node X in the first Bayesian network, which is obtained according to the parent node of node X in the first Bayesian network and the first conditional probability table.

[0085] For example, in the first machine, the edge probability of node X is P1(X)=0.8, while in the second machine, the edge probability of node X is P2(X)=0.6, then the edge probability change of node X is ΔP(X)=0.8-0.6=0.2. This result shows that the probability of node X occurring in the first machine is higher than that in the second machine, indicating that the state of node X in the first machine is more stable or common.

[0086] Step 2: Based on the first conditional probability of each node in the first Bayesian network relative to any parent node and the second conditional probability of the corresponding node in the second Bayesian network relative to the corresponding parent node, determine the change in conditional probability of each node relative to the parent node.

[0087] Specifically, the change in conditional probability of each node relative to any parent node is determined by the following formula:

[0088] in, is the conditional probability change of node X relative to its parent node Y, is the conditional probability of node X given the state of parent node Y in the first machine, is the conditional probability of node X given the state of parent node Y in the second machine.

[0089] For example, in the first machine, given that the state of the parent node Y is 1, the conditional probability that the state of the node X is 1 is =0.9. In the second machine, given that the parent node Y is in state 1, the conditional probability that the node X is in state 1 is =0.7, then the conditional probability change of node X relative to its parent node Y is =0.9-0.7=0.2. This result shows that in the first machine, when the parent node Y is in state 1, the probability that the node X is in state 1 is higher than that in the second machine. This shows that the causal relationship is different in the two machines, and the first machine has a stronger dependence on this causal relationship.

[0090] Step three, the sum of the conditional probability changes of each node relative to all parent nodes is determined as the causal relationship change of each node.

[0091] Specifically, the change in causal relationship of each node can be determined by the following formula:

[0092] Among them, ΔC(X) is the change in the causal relationship of the node, X is the node, Y is the parent node, Pa(X) is the parent node set of node X, is the conditional probability change of node X relative to its parent node Y. ΔC(X) can be used to characterize the causal relationship change of node X before and after the failure.

[0093] Through the above scheme, by comparing the changes of the first Bayesian network and the second Bayesian network before and after the fault, the propagation path of the fault can be deeply analyzed, and the changes in the causal relationship between the variables before and after the fault can be quantified, so as to facilitate inferring the root cause of the fault by analyzing the changes in the causal relationship between different variables.

[0094] Step 103: Determine the root cause score of each node based on the edge probability change and causal relationship change of each node and a preset weighting coefficient.

[0095] Specifically, the root cause score can be used to characterize the role of the node in the device failure.

[0096] In some examples, the preset weighting coefficient includes a first weighting coefficient and a second weighting coefficient; the sum of the first weighting coefficient and the second weighting coefficient is 1. Exemplarily, the first weighting coefficient may be 0.3-0.4, and the second weighting coefficient may be 0.7-0.6. For example, the first weighting coefficient may be set to 0.3, and the second weighting coefficient may be set to 0.7.

[0097] Specifically, the root cause score of each node can be determined by the following formula:

[0098] Where S(X) is the root cause score of node X, W1 is the first weight coefficient, W2 is the second weight coefficient, W1+W2=1; ΔP(X) is the change in edge probability of node X, and ΔC(X) is the change in causal relationship of node X.

[0099] Step 104: Determine the key parameter corresponding to the node with the highest root cause score among all nodes as the fault root cause parameter.

[0100] For example, please refer to Figure 2 and Figure 3, assuming that in the second Bayesian network, node f has the highest root cause score, indicating that the key parameter corresponding to node f plays a core role in the equipment failure. By calculating ΔC(f), it can also be found that the causal relationship of node f changes significantly before and after the failure, and it is inferred that the key parameter corresponding to node f is the root cause of the abnormality of other key parameters.

[0101] It is understandable that the embodiments of the present application have been put into use in silicon wafer cutting production lines. In practical applications, they have helped factories successfully identify the root causes of many complex faults, and can reduce the frequency of similar faults through targeted maintenance. The Bayesian inference method of the embodiments of the present application can deeply analyze the propagation path of the fault by comparing the changes in the DAG before and after the fault, and explore the root cause of the equipment failure. Specifically, the root cause score can be calculated by quantifying the changes in the causal relationship between the variables before and after the fault, and finally the key variables that cause the fault can be located. This root cause location method based on Bayesian inference can not only shorten the troubleshooting time, but also greatly improve the accuracy of fault location. Compared with traditional empirical judgment and troubleshooting methods, the method of the embodiments of the present application is more scientific and efficient, and provides strong technical support for the intelligent maintenance of factories.

[0102] Accordingly, see Figure 4 , Figure 4 Schematic diagram of the structure of the fault root cause determination device of the embodiment of the present application. The fault root cause determination device provided in the embodiment of the present application includes a network construction module 401, a network comparison module 402, a score calculation module 403 and a root cause determination module 404, the network comparison module 402 is connected to the network construction module 401, the score calculation module 403 is connected to the network comparison module 402, and the root cause determination module 404 is connected to the score calculation module 403. Exemplarily, the modules can be connected in communication.

[0103] The network construction module 401 is used to construct a first Bayesian network based on the key parameters of the first machine and a second Bayesian network based on the key parameters of the second machine, the abnormality level of the first machine meets the preset conditions, and the abnormality level of the second machine exceeds the preset conditions.

[0104] The network comparison module 402 is used to determine the edge probability change and causal relationship change of each node based on the first edge probability and the first conditional probability of each node in the first Bayesian network and the second edge probability and the second conditional probability of each node in the second Bayesian network.

[0105] The score calculation module 403 is used to determine the root cause score of each node based on the edge probability change amount and the causal relationship change amount of each node and a preset weighting coefficient.

[0106] The root cause determination module 404 is configured to determine the key parameter corresponding to the node with the highest root cause score among all nodes as the fault root cause parameter.

[0107] In some embodiments, the network comparison module 402 is specifically used to: Based on the first edge probability of each node in the first Bayesian network and the second edge probability of the corresponding node in the second Bayesian network, the edge probability change of each node is determined.

[0108] Based on the first conditional probability of each node in the first Bayesian network relative to any parent node and the second conditional probability of the corresponding node in the second Bayesian network relative to the corresponding parent node, the conditional probability change of each node relative to the parent node is determined.

[0109] The sum of the conditional probability changes of each node relative to all parent nodes is determined as the causal relationship change of each node.

[0110] In some embodiments, the network comparison module 402 is specifically used to: The causal relationship change amount of each node is determined by the following formula:

[0111] Among them, ΔC(X) is the change in the causal relationship of the node, X is the node, Y is the parent node, Pa(X) is the parent node set of node X, is the conditional probability change of node X relative to its parent node Y.

[0112] In some embodiments, the preset weighting coefficient includes a first weighting coefficient and a second weighting coefficient. The score calculation module 403 is specifically used for: The root cause score of each node is determined by the following formula:

[0113] Where S(X) is the root cause score of node X, W1 is the first weight coefficient, W2 is the second weight coefficient, and W1+W2=1. ΔP(X) is the change in edge probability of node X, and ΔC(X) is the change in causal relationship of node X.

[0114] In some embodiments, the first weight coefficient is 0.3 and the second weight coefficient is 0.7.

[0115] In some embodiments, the network construction module 401 is specifically used to: The historical data of each key parameter of the first machine are obtained, and the historical data are divided into a training set and a test set.

[0116] Based on the training set, the maximum expectation method is used to determine the structure of the first Bayesian network and the direction of each edge.

[0117] The maximum likelihood estimation method is used to determine the first conditional probability table of each node in the first Bayesian network.

[0118] The first Bayesian network is verified based on the test set, and the structure of the first Bayesian network and the direction of each edge are modified according to the verification result to obtain the trained first Bayesian network.

[0119] In some embodiments, the network construction module 401 is further configured to: Determine the abnormality rate and abnormality rate variance of each operating machine in a preset time period.

[0120] The operating machine whose abnormality rate is less than the first threshold and whose abnormality variance is less than the second threshold is determined as the first machine.

[0121] The operating machine whose abnormality rate is greater than or equal to the first threshold and / or whose abnormality variance is greater than or equal to the second threshold is determined as the second machine.

[0122] In some embodiments, the network construction module 401 is further configured to: Determine the equipment parameters of each critical control point of the operating machine. The critical control points are used to characterize the equipment points that can affect the process.

[0123] Select key parameters from the equipment parameters of each key control point.

[0124] The parameter detection method of the embodiment of the present application is exemplarily described below.

[0125] Taking the key parameters of the equipment as the left tension arm tension, cutting fluid temperature, cutting fluid flow, wire mesh speed and feed speed as an example, the left tension arm tension is detected by setting tension sensors at the upper wire end and the lower wire segment, and displayed through the interactive interface of the wire cutting equipment; the cutting fluid temperature is detected by setting a temperature sensor at the outlet of the flow pipe to detect the temperature of the outflowing coolant, and displayed through the interactive interface of the wire cutting equipment; the cutting fluid flow is detected by a flow meter installed on the wire cutting equipment, and displayed through the interactive interface of the wire cutting equipment; the wire mesh speed and feed speed are detected by monitoring the rotation speed of the corresponding motor, and displayed through the interactive interface of the wire cutting equipment.

[0126] Wire cutting equipment usually has a variety of sensors or detection devices built in, and the key parameters of the equipment can be directly displayed on the interactive interface of the wire cutting equipment for easy acquisition.

[0127] It can be understood that the fault root cause determination device of the embodiment of the present application determines the fault root cause parameters by comparing the Bayesian networks of machines with different abnormality levels, which can not only shorten the fault investigation time and improve the production efficiency of the production line, but also greatly improve the accuracy of fault location.

[0128] Accordingly, see Figure 5 , Figure 5 Schematic diagram of the structure of an electronic device in an embodiment of the present application. An electronic device provided in an embodiment of the present application includes a memory 501 and a processor 502. The memory 501 is used to store a computer program. The processor 502 is used to execute the computer program stored in the memory 501. When the computer program stored in the memory 501 is executed, the processor 502 executes the method for determining the root cause of a fault in the aforementioned embodiment of the present application.

[0129] Accordingly, an embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to implement the method for determining the root cause of a fault as described in the aforementioned embodiment of the present application when executed.

[0130] The above is a detailed introduction to a method, device and electronic device for determining the root cause of a fault provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the technical solution and its core idea of ​​the present application. Ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents; and these modifications or replacements do not make the essence of the corresponding technical solution deviate from the scope of the technical solution of the embodiments of the present application.

Claims

1. A method for determining a root cause of a fault, characterized in that: include: A first Bayesian network is constructed based on key parameters of a first machine, and a second Bayesian network is constructed based on key parameters of a second machine, wherein the abnormality of the first machine satisfies a preset condition, and the abnormality of the second machine exceeds the preset condition; Determine an edge probability change and a causal relationship change of each node based on a first edge probability and a first conditional probability of each node in the first Bayesian network and a second edge probability and a second conditional probability of each node in the second Bayesian network; Determine the root cause score of each node based on the edge probability change and the causal relationship change of each node and a preset weighting coefficient; The key parameter corresponding to the node with the highest root cause score among all the nodes is determined as the fault root cause parameter.

2. The method for determining the root cause of a fault according to claim 1, characterized in that: The step of determining the edge probability change amount and the causal relationship change amount of each node comprises: Determining an amount of change in the edge probability of each node based on a first edge probability of each node in the first Bayesian network and a second edge probability of a corresponding node in the second Bayesian network; Determine a change in the conditional probability of each node relative to the parent node based on a first conditional probability of each node in the first Bayesian network relative to any parent node and a second conditional probability of the corresponding node in the second Bayesian network relative to the corresponding parent node; The sum of the conditional probability changes of each node relative to all the parent nodes is determined as the causal relationship change of each node.

3. The method for determining the root cause of a fault according to claim 2, characterized in that: The step of determining the causal relationship change amount of each node comprises: The causal relationship change of each node is determined by the following formula: Wherein, ΔC(X) is the change in the causal relationship of the node, X is the node, Y is the parent node, Pa(X) is the parent node set of node X, is the conditional probability change of the node X relative to the parent node Y.

4. The method for determining the root cause of a fault according to claim 1, characterized in that: The preset weighting coefficient includes a first weighting coefficient and a second weighting coefficient; the step of determining the root cause score of each node includes: The root cause score of each of the nodes is determined by the following formula: Wherein, S(X) is the root cause score of node X, W1 is the first weight coefficient, W2 is the second weight coefficient, W1+W2=1; ΔP(X) is the edge probability change of node X, and ΔC(X) is the causal relationship change of node X.

5. The method for determining the root cause of a fault according to claim 4, characterized in that: The first weight coefficient is 0.3, and the second weight coefficient is 0.

7.

6. The method for determining the root cause of a fault according to claim 1, characterized in that: The step of constructing a first Bayesian network based on the key parameters of the first machine includes: Acquire historical data of each of the key parameters of the first machine, and divide the historical data into a training set and a test set; Based on the training set, using the maximum expectation method to determine the structure of the first Bayesian network and the direction of each edge; Determine a first conditional probability table for each node in the first Bayesian network using a maximum likelihood estimation method; The first Bayesian network is verified based on the test set, and the structure of the first Bayesian network and the direction of each edge are corrected according to the verification result to obtain the trained first Bayesian network.

7. The method for determining the root cause of a fault according to claim 1, characterized in that: The first machine and the second machine are determined by the following steps: Determine the abnormality rate and abnormality rate variance of each operating machine in a preset period of time; Determine the operating machine whose abnormality rate is less than the first threshold and whose abnormality rate variance is less than the second threshold as the first machine; The operating machine whose abnormality rate is greater than or equal to the first threshold and / or whose abnormality rate variance is greater than or equal to the second threshold is determined as the second machine.

8. The method for determining the root cause of a fault according to claim 1, characterized in that: The step of determining the key parameters comprises: Determine the equipment parameters of each critical control point of the operating machine, wherein the critical control point is used to characterize the equipment point that can affect the process; The key parameters are selected from the equipment parameters of each of the key control points.

9. A device for determining a root cause of a fault, characterized in that: include: A network construction module, configured to construct a first Bayesian network based on key parameters of a first machine, and to construct a second Bayesian network based on key parameters of a second machine, wherein the abnormality of the first machine satisfies a preset condition, and the abnormality of the second machine exceeds the preset condition; A network comparison module, for determining an edge probability change amount and a causal relationship change amount of each node based on a first edge probability and a first conditional probability of each node in the first Bayesian network, and a second edge probability and a second conditional probability of each node in the second Bayesian network; A score calculation module, used to determine the root cause score of each node based on the edge probability change and the causal relationship change of each node, and a preset weighting coefficient; The root cause determination module is used to determine the key parameter corresponding to the node with the highest root cause score among all the nodes as the fault root cause parameter.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory; the memory stores a computer program, and the processor is used to execute the computer program stored in the memory to implement the fault root cause determination method according to any one of claims 1 to 8.

Images