Fault detection method, electronic equipment, computer program product and system

Abstract

The invention discloses a fault detection method, electronic equipment, a computer program product and a system, which are applied to the field of automation. The method comprises the steps that based on fault detection data, a fault classification result is determined through a fault classification model, the fault classification result comprises a predicted fault type to which the fault detection data belongs, and the fault classification model is a trained random forest model; and based on the fault classification result, determining a predicted fault reason corresponding to the fault detection data through an expert system. The prediction capability of the random forest and the reasoning capability of the expert system are combined, so that a powerful fault detection system can be constructed, and not only can high-accuracy prediction be provided, but also explainable and reliable diagnosis results can be provided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation, and in particular to a fault detection method, an electronic device, a computer program product and a system. BACKGROUND

[0002] In the field of industrial production, enterprises often require equipment to run continuously for a long time in order to meet production needs. For example, with the development of technology, the scale of mechanical equipment is expanding, the structure and control part is becoming more complex, the performance index is constantly improving, and the mutual cooperation and integration between equipment are becoming increasingly close, but the working intensity of mechanical equipment is also increasing.

[0003] Such long-time high-load operation increases the risk of failure of modern equipment, especially large and complex equipment, and the failure mode is more complex and variable. In the face of this challenge, existing fault diagnosis methods have been unable to meet the diagnosis needs of modern equipment, especially large and complex equipment.

[0004] Existing fault diagnosis methods include signal analysis methods, deep learning methods, and expert system-based methods.

[0005] Signal analysis methods are limited by the subjective judgment of technicians and have limitations in application scenarios. For example, vibration analysis is very effective for detecting problems such as imbalance, misalignment, bearing damage, and mechanical looseness. This technology captures the vibration signal of the equipment through a vibration sensor (usually an accelerometer, speedometer, or displacement meter), and then analyzes the frequency, amplitude, envelope, waveform changes, etc. of the vibration signal. The effectiveness of this technology depends on many factors, including the quality of the sensor, the installation location, the signal processing technology, the analysis method, the experience of the technician, etc. For example, oil analysis can accurately detect wear metals, contaminants, and other signs of failure by chemically and physically analyzing machine lubricating oil. This technology analyzes the viscosity, particles, moisture, pH, etc. of machine lubricating oil or other fluids to assess the operating condition of the machine and detect potential failures.

[0006] Deep learning methods have too high a dependence on training data, high training costs, and poor model interpretability. Specifically, deep learning collects data from rotating machine equipment under normal operation and different fault conditions, extracts statistical quantities, frequency domain features, time domain features, etc. from the original data for model training, and finally deploys the trained deep learning model to a real-time monitoring system. This technology has strong capabilities in processing large amounts of data, automatic feature extraction, high accuracy, and adaptability to complex patterns, but also faces many challenges, such as the need for a large amount of labeled data, high computing resource requirements, model interpretability issues, sensitivity to initial settings, etc.

[0007] The traditional method of detecting equipment failure by using an expert system completely relies on a knowledge base and is difficult to handle nonlinear relationships of equipment data and is not flexible. SUMMARY

[0008] The present application aims to provide a failure detection method and system with certain reliability, interpretability and low training cost.

[0009] In a first aspect, the present application provides a failure detection method, comprising: S1, obtaining failure detection data; S2, determining feature values of a plurality of features based on the failure detection data; S3, inputting the feature values of the plurality of features into a failure classification model to obtain a failure classification result, the failure classification result including a predicted failure type to which the failure detection data belongs, the failure classification model being a trained random forest model; and S4, determining a predicted failure cause corresponding to the failure detection data by an expert system based on the failure classification result.

[0010] In some embodiments, the failure classification result further includes at least one key feature in the plurality of features, the at least one key feature being determined based on a ranking result of importance scores of the plurality of features, the importance score of each feature reflecting the importance of the feature to the classification process of the failure detection data.

[0011] In some embodiments, step S4 further comprises: S41, determining one or more candidate failure causes corresponding to the failure detection data by an expert system based on the failure classification result; if a plurality of candidate failure causes are determined, performing step S42 and step S43; otherwise, performing step S44; S42, ranking the candidate failure causes; S43, determining the predicted failure cause based on the ranking result; and S44, determining the only determined candidate failure cause as the predicted failure cause.

[0012] In some embodiments, step S42 further comprises: S421, for each of a plurality of preset scoring strategies, scoring the plurality of candidate failure causes according to the scoring strategy to obtain a special score of the plurality of candidate failure causes under the scoring strategy; S422, for each of the plurality of candidate failure causes, performing weighted summation on the special scores of the candidate failure cause under the plurality of preset strategies to obtain a comprehensive score of the failure cause; and S423, ranking the plurality of candidate failure causes according to the comprehensive scores of the plurality of candidate failure causes.

[0013] In some embodiments, the plurality of preset scoring strategies include any combination of the following strategies:

[0014] According to a targeted scoring strategy, the association degree between the candidate fault cause and the fault state reflected by the fault detection data is determined according to historical data, expert experience or statistical analysis results based on historical data, and the score of the candidate fault cause is determined according to the association degree;

[0015] According to a matching degree scoring strategy, the semantic similarity between the candidate fault cause and the fault state reflected by the fault detection data is determined, and the score of the candidate fault cause is determined according to the semantic similarity;

[0016] According to a working environment scoring strategy, the score of the candidate fault cause is determined according to the consistency between the condition met by the working environment reflected by the fault detection data and the establishment condition of the fault cause;

[0017] According to a time information scoring strategy, the score of the candidate fault cause is determined according to the time information in the fault detection data.

[0018] In some embodiments, the method is applied to fault detection of a rotating device, and the fault detection data includes at least one of sensor data, device operation parameter data, historical maintenance data and a reference fault data set.

[0019] In some embodiments, the sensor data includes at least one of vibration data, temperature data, voltage data and current data, and the device operation parameter data includes at least one of operation speed, operation load and continuous operation time length.

[0020] In a second aspect, the present application provides a fault detection system, comprising: an acquisition module configured to acquire fault detection data; a determination module configured to determine feature values of a plurality of features based on the fault detection data; a classification module configured to: input the feature values of the plurality of features into a fault classification model to obtain a fault classification result, the fault classification result including a predicted fault type to which the fault detection data belongs, the fault classification model being a trained random forest model; and an explanation module configured to determine, based on the fault classification result, a predicted fault cause corresponding to the fault detection data through an expert system.

[0021] In a third aspect, the present application provides an electronic device, comprising a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, the fault detection method of the first aspect is implemented.

[0022] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and when a computer executes the computer program, the fault detection method of the first aspect is implemented.

[0023] Fifthly, the present invention provides a computer program product that, when run on a computer, causes the computer to perform the fault detection method as described in the first aspect.

[0024] In this invention, random forests excel at handling large amounts of data and discovering complex patterns (such as nonlinear relationships), while expert systems are superior in handling rule-based reasoning. Combining the two allows for complementary advantages, including accuracy, flexibility, and adaptive learning capabilities. By combining the predictive power of random forests with the reasoning power of expert systems, this invention can construct a powerful fault detection system that not only provides highly accurate predictions but also interpretable and reliable diagnostic results. Attached Figure Description

[0025] The accompanying drawings, which are provided to further illustrate the invention and constitute a part of this invention, are not intended to limit the invention.

[0026] Figure 1 This is an exemplary flowchart of the fault detection method provided in the embodiments of the present invention.

[0027] Figure 2 This is an exemplary structural diagram of the integrated model provided in the embodiments of the present invention.

[0028] Figure 3 This is a schematic diagram of a semantic network.

[0029] Figure 4 This is a schematic diagram of the framework.

[0030] Figure 5 This is an exemplary flowchart provided by an embodiment of the present invention, which uses an expert system to determine the predicted cause of a fault corresponding to fault detection data.

[0031] Figure 6 This is an exemplary flowchart provided by an embodiment of the present invention for sorting the multiple candidate fault causes according to multiple scoring strategies.

[0032] Figure 7 This is a schematic diagram of the fault detection process based on an integrated model provided in an embodiment of the present invention.

[0033] Figure 8 This is an exemplary module diagram of the image decompression system provided in the embodiments of the present invention.

[0034] Figure 9 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0035] To better understand the technical content and advantages of the present invention, the present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. The present invention will now be described in further detail with reference to the accompanying drawings.

[0036] Figure 1 These are exemplary flowcharts illustrating some embodiments of this specification. Figure 1 As shown, process 100 includes steps S1 to S3.

[0037] S1, acquire fault detection data.

[0038] Fault detection data can reflect the status of equipment, such as abnormal (fault) conditions. In some embodiments, fault detection data includes at least one of sensor data, equipment operating parameter data, historical maintenance data, and reference fault datasets.

[0039] By installing sensors at appropriate locations on the equipment, the required data, i.e., sensor data, can be collected. These sensors include, but are not limited to, vibration sensors, temperature sensors, and current / voltage sensors. Accordingly, in some embodiments, the sensor data includes at least one of vibration data, temperature data, voltage data, and current data. Different types of sensors can be used for different equipment; for example, vibration / temperature sensors are recommended for machine tool spindle equipment, and voltage / current sensors are recommended for generator equipment.

[0040] For example, equipment operating parameter data can be obtained from the equipment's dashboard or control system. In some embodiments, the equipment operating parameter data includes at least one of the following: operating speed, operating load, continuous operating duration, etc.

[0041] Historical maintenance data can include components with maintenance records and their maintenance records, such as the repair (or maintenance) time and repair (or maintenance) items for a certain component.

[0042] For example, the reference fault dataset can be an open data source, such as the dataset published by the Intelligent Diagnosis and Expert System (IDES) Laboratory of Nanjing University of Aeronautics and Astronautics, which includes various fault simulation data of rolling bearings, such as fault simulation datasets for rolling bearings of different sizes.

[0043] In some embodiments, the fault detection method is applied to fault detection of rotating equipment.

[0044] S2, determine the feature values ​​of multiple features based on the fault detection data.

[0045] Different feature values ​​can be used to represent different states of a device; in short, it maps the device state to numerical values. Taking vibration data collected by a vibration sensor as an example, the mean, variance, and spectrum can be calculated as features based on the vibration data.

[0046] In some embodiments, the plurality of features includes one or more features selected from time-domain features, frequency-domain features, and time-frequency-domain features. For example, frequency-domain feature extraction may include calculating the mean and standard deviation for each time window, where a high standard deviation indicates possible abnormal vibration in the device; time-domain feature extraction may include converting the time-domain signal into frequency-domain features using Fourier transform and analyzing the power spectral density of the time-domain signal (such as a vibration signal), where abnormal frequency components indicate a possible device malfunction; time-frequency-domain feature extraction may include combining time signals and frequency information, such as wavelet transform, to decompose the signal into different frequency components and analyze them in the time dimension.

[0047] S3, input the feature values ​​of the multiple features into the fault classification model to obtain the fault classification result.

[0048] The fault classification results include the predicted fault type to which the fault detection data belongs. Taking rotating equipment as an example, the label space of the fault classification model can include bearing wear, gear failure, balance failure, and no fault. The predicted fault type to which the fault detection data belongs is one of these preset fault types.

[0049] The fault classification model is a trained random forest model (or simply random forest). A random forest is an ensemble model that comprises multiple decision trees; its output is the combination of the predictions from these decision trees. Each tree is built independently, but they share the same objective function. When training each decision tree, samples are drawn from the original training set using a bootstrap sampling method. This means that the same sample may be drawn multiple times, and each tree is trained using a different set of samples. Because it integrates multiple decision trees, random forests typically offer high prediction accuracy, and due to the diversity of the ensemble model reducing the impact of individual outliers, they exhibit good robustness.

[0050] The training set for the fault classification model can be obtained from the original dataset. More details about the original data can be found in the documentation for fault detection data. For example, the original data may include at least one of the following: sensor data, equipment operating parameter data, historical maintenance data, and reference fault datasets.

[0051] To obtain a training set, the original data can be preprocessed, and features can be extracted from the preprocessed data.

[0052] For example, preprocessing includes data standardization, handling imbalanced data, and data segmentation. Standardization refers to adjusting data to a uniform scale or range to eliminate the influence of different units / magnitudes. Common methods include min-max method, z-score method, and scaling method. Data cleaning refers to removing erroneous, duplicate, or inconsistent data records from the dataset. For example, if sensor data is lost due to signal loss or other malfunctions, data filling can be chosen; if a large amount of data is missing for a certain period, this part of the data can be discarded to avoid affecting the overall analysis; or if the recorded data is incorrect, such as the specification stating that the rotation speed should not exceed a certain threshold, but the recorded parameter clearly exceeds this value, this data needs to be removed. Data segmentation refers to dividing a continuous data stream into fixed-length time periods or event-based data segments. For example, given a year's worth of sensor data, it can be divided chronologically into a 70% training set (data from January to August), a 15% validation set (data from September to October), and a 15% test set (data from November to December). This time-series segmentation method ensures that the model learns from past data during training and evaluates future performance during validation and testing.

[0053] To obtain the training set, the labels (true categories) of each sample can also be annotated (e.g., manually). That is, each sample includes feature values ​​and labels for the multiple features.

[0054] A training set was constructed based on the aforementioned data and its corresponding feature set. For random forests, samples can be drawn from the training set using bootstrap sampling to train each decision tree, meaning that each sample may be selected repeatedly. The training of the decision tree employs methods such as information gain, Gini impurity, and gain ratio to select the optimal feature for splitting. Considering the model's interpretability, the gain ratio method is chosen to select the optimal feature. This balances the feature selection by considering the feature's ability to segment data (information gain) and the complexity of the segmentation (segmentation entropy), avoiding the selection of high-branch features that may lead to overfitting.

[0055] The formula for calculating the gain ratio is as follows:

[0056]

[0057] Where H(D) represents the information entropy of dataset D, H(D|A) represents the conditional information entropy corresponding to feature A, and H(D)-H(D|A) represents the information gain; H A (D) represents the splitting entropy.

[0058] The formula for calculating information entropy H(D) is as follows:

[0059]

[0060] Among them, C k Let |D| represent the number of samples belonging to class k, and let |D| represent the number of samples in D. The total number of samples in all classes is equal to the total number of samples in dataset D.

[0061] The formula for calculating the conditional information entropy H(D|A) is as follows:

[0062]

[0063] Where n is the number of all possible values ​​of feature A (i.e., feature A has a total of n possible values), D i Let represent the set of samples belonging to the i-th category.

[0064] Splitting Entropy H A The formula for calculating (D) is as follows:

[0065]

[0066] A higher gain ratio indicates less uncertainty in the features, making it more capable of splitting effective nodes in the decision tree, which helps improve the model's prediction accuracy.

[0067] In some embodiments, the fault classification result further includes at least one key feature from the plurality of features, which is determined based on the ranking of the importance scores of the plurality of features. The importance score of each feature reflects its importance to the classification process of the fault detection data. Each decision tree in the random forest performs feature matching on the input data. The random forest integrates the matching results using a probability averaging method to obtain the feature that has the greatest impact on the prediction result, i.e., the key feature. Because the random forest can evaluate the contribution of different features to fault classification and provide the expert system with insights into which features are key fault indicators, the prediction result of the random forest is used as the input to the expert system. The input here includes two parts: one is the fault type ultimately predicted by the random forest, and the other is the features that have a significant impact on fault classification, determined based on the feature importance scores. These features can help the expert system understand which factors play a key role in fault prediction.

[0068] S4. Based on the fault classification results, the predicted fault causes corresponding to the fault detection data are determined through an expert system.

[0069] Random forests excel at handling non-linear relationships in data, helping to improve the accuracy of ensemble models (combining random forests and expert systems). They also possess strong generalization capabilities, providing expert systems with the most relevant outcome predictions by evaluating the importance of features for fault classification. Expert systems, on the other hand, are highly interpretable, possessing logical reasoning abilities and capable of deriving possible causes of faults based on symptoms and conditions. By fully leveraging the capabilities of random forests in handling complex datasets while utilizing the strengths of expert systems in providing decision support and interpretation, more efficient and reliable fault detection can be achieved.

[0070] An expert system is a computer program system that simulates human experts in solving problems in a specific domain. Figure 2 This is an exemplary structural diagram of the integrated model provided in an embodiment of the present invention. (Reference) Figure 2 An expert system includes an inference engine, a knowledge base, a comprehensive database, and a human-computer interface.

[0071] Human-computer interfaces (HCIs) serve as a medium for information exchange between humans and expert systems. For example, users, experts, and engineers can access expert systems through HCIs. HCIs can not only recognize and interpret commands, questions, and data provided by users to the system and transform this information into the system's internal representation, but also provide users with the questions posed by the expert system, the results obtained, and the explanations offered in a form easily understood by the user.

[0072] A comprehensive database is a database (or working memory) used by an expert system to store various information needed and generated during the reasoning process. The data in the comprehensive database is primarily used to describe the problem state. It can store static information about devices, historical maintenance records, operating parameters, historical fault cases, etc., and this data can be organized into various forms such as text descriptions, time-series data, and parameter configurations. The comprehensive database can also be used for initial state descriptions of the problem to be solved, intermediate results, records of the solution process, and questions and answers between the user and the system.

[0073] The knowledge base is the core component of an expert system. It stores knowledge related to equipment fault detection, including expert interviews, questionnaires, research literature, and case studies. The knowledge in the knowledge base supports classification, organization, and retrieval, and can be continuously iterated and updated. Knowledge in the knowledge base can be represented in various forms, including but not limited to production rules (rules), cases, semantic networks, and frames. A rule consists of two parts: a condition and a conclusion; that is, satisfying a certain condition yields a corresponding conclusion. The format of a rule can be IF [condition] THEN [conclusion].

[0074] Case studies are historical cases. By searching for similar, solved cases from the past, comparing the characteristics and conditions under which the new and old problems occur, we can identify similarities and differences. Then, by referencing existing knowledge, we can reason again and arrive at new methods for solving the problems.

[0075] Semantic networks graphically represent concepts (or entities) and the relationships between them. (See reference) Figure 3 In semantic networks, nodes represent concepts or entities, while edges represent the semantic relationships or associations between them. Nodes can be different parts of a device (such as bearings, gears, motors, etc.) and failure modes (such as overheating, vibration, imbalance, etc.), while edges can represent assembly relationships between components, causal relationships between failures, or associations between failures and potential causes.

[0076] refer to Figure 4 Each frame can represent a component of the equipment (e.g., an electric motor or a bearing) and can be used to store the component's type, parameters (e.g., power, material), fault type, diagnostic methods, etc. Frames within and between frames are connected by slots used to store dimensions, materials, installation information, performance parameters, historical maintenance records, etc.

[0077] Inference engines can deduce conclusions based on knowledge. Inference mechanisms generally fall into two categories: forward reasoning and backward reasoning. Forward reasoning starts with known facts (considered conditions) and applies rules from a knowledge base to progressively deduce new conclusions. Backward reasoning starts with the target state or problem (considered a conclusion) and applies rules in reverse to identify the necessary conditions or prerequisites. Forward reasoning is used when monitoring equipment status under normal operating conditions, while backward reasoning is used when equipment malfunctions and requires verification and resolution. Here, we combine forward and backward reasoning, continuously updating the knowledge base rules based on subsequent real-world failures and their solutions to monitor and detect equipment status. This hybrid reasoning also requires a conflict resolution logic; when the two reasoning approaches lead to contradictory conclusions, a choice must be made.

[0078] The knowledge base contains a large number of rules. By matching the corresponding rules, the inference engine can obtain possible causes of failure. This can lead to a situation where a single device failure symptom corresponds to multiple conflicting causes. In some embodiments, a sorting strategy can be employed to resolve potential conflicts. Several exemplary sorting strategies are described below.

[0079] The targeted sorting strategy refers to ranking causes based on their correlation with fault symptoms. This correlation can be based on historical data, expert experience, or statistical analysis results from similar situations. The expert system evaluates the degree of match between each cause and the current fault symptom, ranking the cause with the highest match (i.e., the most likely cause of the current fault) first. Taking abnormal vibration faults as an example, the most likely causes are bearing wear and rotor imbalance, followed by loose mechanical parts and gear wear, and then motor failures (overheating or insulation damage, etc.). Starting with the most likely cause, other secondary causes are gradually eliminated. This orderly elimination process not only saves time but also improves diagnostic accuracy, avoiding additional costs and equipment downtime caused by misdiagnosis.

[0080] The strategy of ranking by match degree is similar to that of ranking by specificity, but it places greater emphasis on the accuracy of the match between the cause and the symptom in terms of rules or patterns. For example, if the description of a cause perfectly matches the symptom of the fault, it will receive a higher match degree score. Taking the fault phenomenon of abnormal noise from a machine accompanied by increased vibration as an example, the highest match degree is for a damaged rolling bearing, because it causes abnormal noise and vibration during machine rotation, usually resulting in mechanical vibration and noticeable noise, especially when there is wear or debris inside the bearing. The following causes, ranked from highest to lowest match degree, are rotor imbalance, damaged conveyor belts, and poor gear meshing. These causes are usually accompanied by other special circumstances, so their match degree is not as high. For example, rotor imbalance is usually accompanied by uneven wear, the noise from a damaged conveyor belt is generally specific, and poor gear meshing is usually an impact sound that is periodic and not continuous.

[0081] A strategy that prioritizes causes based on contextual constraints (or operating environment) considers the impact of the current context or specific conditions on the cause of a failure. For example, if a cause only occurs under specific operating conditions, and the current environment meets those conditions, then that cause will be given higher priority.

[0082] The strategy of sorting by the freshness of known facts (or time information) refers to sorting based on the timestamps of known facts (such as recent sensor readings or user reports). Newer data may better reflect the current state of the system, so relevant causes will be given priority.

[0083] In some embodiments, the above sorting methods can be combined to obtain a reliable sorting method through weighted summation. For more details on sorting fault causes, please refer to [reference needed]. Figure 5 , Figure 6 And its related descriptions.

[0084] Figure 5This is an exemplary flowchart provided by an embodiment of the present invention, showing how an expert system determines the predicted cause of a fault corresponding to fault detection data (i.e., step S4). Figure 5 As shown, in some embodiments, step S4 further includes the following steps.

[0085] S41, based on the fault classification results, determine one or more candidate fault causes corresponding to the fault detection data through an expert system.

[0086] The explanation module 240 can determine one or more candidate fault causes corresponding to the fault detection data by retrieving knowledge (e.g., cases and rules) from the knowledge base.

[0087] If multiple candidate causes of failure are identified, the interpretation module 240 executes steps S42 and S43. Otherwise (i.e., only one candidate cause of failure is identified), the interpretation module 240 executes step S44.

[0088] S42, sort the multiple candidate fault causes.

[0089] In some embodiments, the interpretation module 240 sorts the plurality of candidate fault causes according to multiple scoring strategies, as detailed below. Figure 3 And its related descriptions.

[0090] S43, determine the predicted cause of failure based on the sorting results.

[0091] The interpretation module 240 can select at least one candidate fault cause as the predicted fault cause based on the ranking results. For example, the interpretation module 240 can determine the candidate fault cause ranked first as the predicted fault cause.

[0092] S44 identifies the uniquely determined candidate cause of failure as the predicted cause of failure.

[0093] Figure 6 This is an exemplary flowchart of an embodiment of the present invention, which sorts the multiple candidate fault causes according to multiple scoring strategies (i.e., step S42). Figure 6 As shown, step S42 further includes steps S421 to S423.

[0094] S421, For each of the multiple preset scoring strategies, score the multiple candidate fault causes according to the scoring strategy to obtain the specific score of the multiple candidate fault causes under the scoring strategy.

[0095] S422, for each of the plurality of candidate fault causes, the specific scores of the candidate fault cause under the plurality of preset strategies are weighted and summed to obtain the comprehensive score of the fault cause.

[0096] The weights corresponding to each scoring strategy can be set according to actual needs.

[0097] S423, Rank the multiple candidate fault causes according to the comprehensive score of the multiple candidate fault causes.

[0098] The explanation module 240 can sort the multiple candidate fault causes according to the comprehensive score, and the candidate fault cause with the higher the comprehensive score will be ranked higher in the sorting results.

[0099] In some embodiments, the plurality of preset scoring strategies include any combination of the following strategies: a scoring strategy based on targeting, a scoring strategy based on matching degree, a scoring strategy based on working environment, and a scoring strategy based on time information. Specifically, the scoring strategy based on targeting is used to determine the degree of correlation between candidate fault causes and the fault state reflected in fault detection data based on historical data, expert experience, or statistical analysis results based on historical data, and to determine the score of the candidate fault cause based on the degree of correlation; the scoring strategy based on matching degree is used to determine the semantic similarity between candidate fault causes and the fault state reflected in fault detection data, and to determine the score of the candidate fault cause based on the semantic similarity; the scoring strategy based on working environment is used to determine the score of the candidate fault cause based on the consistency between the conditions met by the working environment reflected in the fault detection data and the conditions for the establishment of the fault cause; and the scoring strategy based on time information is used to determine the score of the candidate fault cause based on the time information in the fault detection data.

[0100] Figure 7 This is a schematic diagram of the fault detection process based on an integrated model provided in an embodiment of the present invention. Figure 7 As shown, before building the ensemble model, it is necessary to train the random forest and construct the expert system separately. After obtaining the ensemble model, the random forest can be used to predict the fault type first, and then the expert system can be used to interpret the fault cause. Specifically, historical cases and production rules are first combined to match candidate fault causes. When multiple candidate fault causes are matched, the predicted fault cause can be output by combining the matching degree (50%) and the relevance (50%). For example, assuming that the equipment has problems such as increased vibration, increased noise, and increased temperature, after matching, the matching degree score for bearing wear is 9 points, the relevance score is 8 points, and the final weighted summation score is 8.5 points. If only one fault cause is matched, the matching result can be output directly.

[0101] The knowledge base can be updated and iterated based on customer feedback. For example, customers may report cases of faults that were not accurately diagnosed or problems that were solved in other ways. The fault detection system can collect and analyze this feedback, such as identifying the reasons why the expert system failed to diagnose the fault accurately or the features it missed, and then update the knowledge base based on the analysis results. For example, the system predicts that the fault is caused by bearing wear, but the actual fault is caused by insufficient bearing lubrication. After analysis, it is found that the system gives low weight to the feature of poor lubrication, causing it not to be identified as a primary fault cause. Then, the rules and feature weights in the knowledge base can be updated. Updates may include adding new fault causes, adjusting the priority of existing causes, and modifying the triggering conditions of rules. Finally, the new rules and weights can be tested, making the fault detection system increasingly intelligent and effective over time.

[0102] It should be noted that the above description of the process is for illustrative purposes only and does not limit the scope of the invention. Those skilled in the art can make various modifications and changes to the process under the guidance of this invention.

[0103] Figure 8 This is an exemplary block diagram of the image compression system provided in this embodiment of the invention. The fault detection system 200 (hereinafter referred to as System 200) can be implemented through software, hardware, or a combination of both. Figure 9 At least a portion of the electronic device 300 shown. For example... Figure 8 As shown, system 200 includes an acquisition module 210, a determination module 220, a classification module 230, and an interpretation module 240.

[0104] The acquisition module 210 is used to acquire fault detection data.

[0105] The determination module 220 is used to determine the feature values ​​of multiple features based on the fault detection data.

[0106] The classification module 230 is used to: input the feature values ​​of the multiple features into a fault classification model to obtain a fault classification result. The fault classification result includes the predicted fault type to which the fault detection data belongs, and the fault classification model is a trained random forest model.

[0107] The interpretation module 240 is used to determine the predicted fault cause corresponding to the fault detection data through an expert system based on the fault classification result.

[0108] The fault detection system and fault detection method provided in this embodiment of the invention belong to the same concept. For more details regarding the system and its modules, please refer to... Figure 1 The details and related descriptions will not be repeated here.

[0109] It should be noted that the above division of functional modules is only an example. For those skilled in the art, after understanding the system principle, they can arbitrarily combine, split, or replace the modules, as well as add or omit one or more modules, without violating the system principle.

[0110] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 300 includes a processor 310 and a memory 320. The memory 320 stores a computer program. When the processor 310 executes the computer program, it implements the fault detection method provided in the embodiment of the present invention.

[0111] This invention also provides a computer-readable storage medium storing a computer program. When a computer executes the computer program, it implements the fault detection method provided in this invention.

[0112] This invention also provides a computer program product that, when run on a computer, causes the computer to execute the fault detection method provided in this invention.

[0113] More details about fault detection methods and procedures can be found at [website address]. Figure 1 The relevant descriptions can be found here, so I will not repeat them here.

[0114] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0115] The above provides a detailed description of the fault detection method and system provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A fault detection method, characterized in that, include: S1, acquire fault detection data; S2, determine the feature values ​​of multiple features based on the fault detection data; S3, input the feature values ​​of the multiple features into the fault classification model to obtain the fault classification result. The fault classification result includes the predicted fault type to which the fault detection data belongs. The fault classification model is a trained random forest model. S4. Based on the fault classification results, the predicted fault cause corresponding to the fault detection data is determined by an expert system.

2. The fault detection method as described in claim 1, characterized in that, The fault classification result also includes at least one key feature among the plurality of features. The at least one key feature is determined based on the ranking result of the importance scores of the plurality of features. The importance score of each feature reflects the importance of the feature to the classification process of the fault detection data.

3. The fault detection method as described in claim 1, characterized in that, Step S4 further includes: S41, Based on the fault classification results, one or more candidate fault causes corresponding to the fault detection data are determined by an expert system; If multiple candidate causes of failure are identified, proceed to steps S42 and S43; otherwise, proceed to step S44. S42, sort the candidate fault causes; S43, determine the predicted cause of the fault based on the sorting results; S44, the uniquely identified candidate fault cause is determined as the predicted fault cause.

4. The fault detection method as described in claim 3, characterized in that, Step S42 further includes: S421, For each of the multiple preset scoring strategies, score the multiple candidate fault causes according to the scoring strategy to obtain the specific score of the multiple candidate fault causes under the scoring strategy. S422, For each of the plurality of candidate fault causes, the specific scores of the candidate fault cause under the plurality of preset strategies are weighted and summed to obtain the comprehensive score of the fault cause. S423, Rank the multiple candidate fault causes according to the comprehensive score of the multiple candidate fault causes.

5. The fault detection method as described in claim 4, characterized in that, The multiple preset scoring strategies include any combination of the following strategies: The targeted scoring strategy is used to: determine the degree of correlation between candidate fault causes and fault states reflected by fault detection data based on historical data, expert experience, or statistical analysis results based on historical data, and to determine the score of candidate fault causes based on the degree of correlation. The matching score strategy is used to: determine the semantic similarity between candidate fault causes and fault states reflected by fault detection data, and determine the score of candidate fault causes based on the semantic similarity. The scoring strategy based on the working environment is used to: determine the score of candidate fault causes based on the consistency between the working environment conditions reflected by fault detection data and the conditions for the establishment of fault causes. The scoring strategy based on time information is used to determine the score of candidate fault causes based on the time information in the fault detection data.

6. The fault detection method as described in claim 1, characterized in that, The method is applied to fault detection of rotating equipment, and the fault detection data includes at least one of the following: sensor data, equipment operating parameter data, historical maintenance data, and reference fault dataset.

7. The fault detection method as described in claim 6, characterized in that, The sensor data includes at least one of vibration data, temperature data, voltage data, and current data, and the equipment operating parameter data includes at least one of operating speed, operating load, and continuous working duration.

8. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the fault detection method as described in any one of claims 1 to 7.

9. A computer program product, characterized in that, When the computer program product is run on a computer, the computer performs the fault detection method as described in any one of claims 1 to 7.

10. A fault detection system, characterized in that, The system, applicable to the fault detection method as described in any one of claims 1 to 7, comprises: The acquisition module is used to acquire fault detection data; The determination module is used to determine the feature values ​​of multiple features based on the fault detection data; The classification module is used to: input the feature values ​​of the multiple features into the fault classification model to obtain the fault classification result, wherein the fault classification result includes the predicted fault type to which the fault detection data belongs, and the fault classification model is a trained random forest model; An interpretation module is used to determine the predicted fault cause corresponding to the fault detection data through an expert system based on the fault classification results.

Images