Equipment fault type classification and root cause analysis system based on isolated forest algorithm

By combining isolated forest algorithm, support vector machine and deep learning, and combining adaptive learning module, the traditional fault location method is solved in the accuracy and efficiency of the 5G network environment, and efficient and accurate device fault classification and root cause are achieved. Analysis has improved the intelligence level of the system.

CN119089327AInactive Publication Date: 2024-12-06NINGBO WILL INFORMATION SCI & TECH CO LTD

Patent Information

Application Number
CN202411579944.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-07
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional human-made equipment fault location method is difficult to meet the needs of the rapidly developing 5G communication network environment in terms of root cause positioning accuracy and positioning efficiency. The isolated forest algorithm has limitations in abnormal classification and root cause analysis.

Method used

Combining the isolated forest algorithm, support vector machine and deep learning, through data preprocessing, abnormal detection, classification and root cause analysis modules, the adaptive learning module dynamic adjustment model is adopted to achieve efficient and accurate classification and root cause analysis of equipment failures.

Benefits of technology

It provides a more efficient, accurate and intelligent device fault diagnosis solution, which can quickly detect abnormal points, accurately classify fault types, and deeply analyze the causes of faults, improving the intelligence level and adaptability of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SMS_17
    Figure SMS_17
  • Figure QLYQS_17
    Figure QLYQS_17
Patent Text Reader

Abstract

The invention provides an equipment fault type classification and root cause analysis system based on an isolated forest algorithm, and the system comprises a data collection and preprocessing module, an anomaly detection module, an anomaly classification module, a root cause analysis module, and a self-adaptive learning module. The data collecting and preprocessing module is used for collecting environment data, equipment state data and production result data from an equipment sensor and carrying out cleaning, abnormal point processing and standardization on the environment data, the equipment state data and the production result data; the anomaly detection module detects abnormal points in the production data through an isolated forest algorithm, and the isolated forest algorithm detects abnormal points different from most data points through randomly constructing a binary tree; the anomaly classification module adopts a support vector machine to classify detected anomaly points, and classification fault types comprise equipment faults, environment anomalies, misoperation and the like; the root cause analysis module analyzes possible causes of faults based on an association rule mining technology, and measures the effectiveness of the rule by using support degree, confidence degree and improvement degree.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault location, and in particular to an equipment fault type classification and root cause analysis system based on an isolation forest algorithm. Background Art

[0002] Isolation Forest (iForest) is an unsupervised learning algorithm that is mainly used to identify outliers in data. The algorithm is simple and efficient, suitable for high-dimensional data and large-scale data sets, and is widely used in the industry. ‌With the rapid development of 5G communication technology, the complexity of the network environment and the amount of data have increased dramatically, which has brought unprecedented challenges to the location and analysis of network faults. Traditional manual positioning methods can no longer meet the requirements in terms of root cause positioning result accuracy and positioning efficiency. Although the isolation forest algorithm is good at detecting anomalies, it has limitations in anomaly classification and root cause analysis. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides an equipment fault type classification and root cause analysis system based on the isolation forest algorithm, which combines the isolation forest with other machine learning algorithms (such as support vector machine (SVM) or deep learning) to further classify the detected anomalies and infer possible fault types and causes through historical data analysis. By introducing an adaptive learning module, dynamic adaptation and intelligent optimization of the production environment are achieved, providing a more efficient, accurate and intelligent solution for equipment maintenance and troubleshooting.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: A system for classifying equipment failure types and analyzing root causes based on an isolation forest algorithm, the system comprising: The data collection and preprocessing module is used to collect environmental data, equipment status data, and production result data from equipment sensors, and clean, process abnormal points, and standardize them; An anomaly detection module detects anomalies in production data using an isolation forest algorithm, which detects anomalies that are different from most data points by randomly constructing a binary tree; The abnormality classification module uses support vector machine (SVM) to classify the detected abnormal points. The classified fault types include equipment failure, environmental abnormality and operation error, etc. The root cause analysis module analyzes the possible causes of failures based on association rule mining technology and uses support, confidence and lift to measure the effectiveness of the rules.

[0005] Preferably, the anomaly score of the isolation forest algorithm is calculated by the following formula: in: For sample Average path length in random forests; is the average path length of the binary tree, which depends on the number of samples and is calculated as follows: in For the The number of harmonics.

[0006] Preferably, the classification decision function of the support vector machine is: in: is the weight vector; is the bias term, The symbol decision Belongs to the category.

[0007] Preferably, the root cause analysis module uses association rule mining technology to analyze the cause of the fault according to the following indicators: Support: indicates the frequency of event A and event B occurring simultaneously. The calculation formula is: Confidence: It indicates the probability of event B occurring when event A occurs. The calculation formula is: Lift: Indicates the strength of the association between event A and event B. The calculation formula is:

[0008] Preferably, the root cause analysis module adopts hierarchical association rule mining technology, including: The first level of causal analysis identifies the direct causes of the failure; The second layer of causal analysis is to explore the potential causes of the first layer failures; The third and deeper causal analysis further analyzes the deep causes of failures caused by equipment maintenance, operating behavior or other external factors.

[0009] Preferably, the hierarchical association rule mining technology uses different data sets at different levels, the input data of the first level is the original production data, including sensor data and equipment status data, and the input data of subsequent levels is a combination of the association results output by the previous level and the original data.

[0010] Preferably, as the level goes deeper, the support and confidence thresholds are dynamically adjusted, and the support threshold gradually decreases as the level goes deeper, so as to ensure the excavation of deep fault root causes.

[0011] Preferably, the system includes an adaptive learning module, which updates the classification model and association rules based on changes in real-time data, and uses an online learning mechanism to dynamically adjust the classification boundaries of the support vector machine to ensure that the classification model adapts to changes in the production environment.

[0012] Preferably, the adaptive learning module receives user feedback on the analysis results through a user feedback mechanism, and the feedback results are used to adjust the association rule model of the root cause analysis module to improve the root cause analysis accuracy of the system.

[0013] Preferably, the data collection and preprocessing module includes a sliding window processing function for time series data, through which the data continuously collected by the sensor is processed in segments to generate a data set suitable for association rule mining.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides an equipment fault type classification and root cause analysis system based on an isolation forest algorithm. The system adopts a combination of the isolation forest algorithm and the support vector machine algorithm, so that the system can quickly and accurately detect and classify abnormal points; through preprocessing steps such as data cleaning, abnormal point processing and standardization, as well as accurate abnormal classification and root cause analysis, the system can provide high-accuracy fault diagnosis results.

[0015] The present invention introduces an adaptive learning module, so that the system can be dynamically adjusted and optimized according to changes in the actual production environment, thereby improving the intelligence level of the system.

[0016] Each module of the present invention is designed to be independent and expandable, which is convenient for function expansion and performance optimization according to actual needs. DETAILED DESCRIPTION

[0017] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be used for other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0018] The present invention provides an equipment failure type classification and root cause analysis system based on an isolation forest algorithm, comprising: The data collection and preprocessing module is used to collect environmental data, equipment status data, and production result data from equipment sensors, and clean, process abnormal points, and standardize them; An anomaly detection module detects anomalies in production data using an isolation forest algorithm, which detects anomalies that are different from most data points by randomly constructing a binary tree; The abnormality classification module uses support vector machines to classify the detected abnormal points. The fault types include equipment failure, environmental abnormality, and operational errors. The root cause analysis module analyzes the possible causes of failures based on association rule mining technology and uses support, confidence and lift to measure the effectiveness of the rules; The adaptive learning module updates the classification model and association rules based on the changes in real-time data, and uses the online learning mechanism to dynamically adjust the classification boundaries of the support vector machine to ensure that the classification model adapts to changes in the production environment.

[0019] It should be noted that the data collection and preprocessing module of this application is responsible for collecting various types of data from equipment sensors, including environmental data, equipment status data and production result data. After cleaning, outlier processing and standardization, these data provide a reliable basis for subsequent analysis.

[0020] The application utilizes the isolation forest algorithm to efficiently detect anomalies in production data. The isolation forest algorithm can quickly identify anomalies that are significantly different from most data points by randomly constructing a binary tree, providing a basis for subsequent fault classification and root cause analysis.

[0021] The abnormal classification module of this application uses the support vector machine algorithm to accurately classify the detected abnormal points. The classification results include equipment failure, environmental anomalies, and operational errors, which helps maintenance personnel quickly locate the problem type and take corresponding solutions.

[0022] The root cause analysis module of this application is based on association rule mining technology, which can deeply analyze the possible causes of failures. By calculating indicators such as support, confidence and lift, the effectiveness of association rules is measured to provide maintenance personnel with valuable root cause analysis information.

[0023] The adaptive learning module of this application is the intelligent core of the entire system. It dynamically updates the classification model and association rules based on the changes in real-time data. Through the online learning mechanism, the adaptive learning module can adjust the classification boundaries of the support vector machine in real time to ensure that the classification model always adapts to changes in the production environment. This adaptive capability enables the system to continuously learn and optimize during long-term operation to improve the accuracy of fault detection and classification.

[0024] The following is a detailed description of each module of this application: The data collection and preprocessing module in the present application includes using device sensors for data collection and using an isolation forest algorithm for data preprocessing; The types of data collected include: environmental data (such as temperature, humidity, vibration), equipment status data (such as voltage, current, load, etc.), and production result data (such as defective rate, downtime rate, etc.).

[0025] Data preprocessing: Clean the collected data, process outliers, standardize and discretize the time series data. The time series data can be discretized using a sliding window method to facilitate association rule analysis.

[0026] The isolation forest algorithm is used to process outliers. The isolation forest algorithm detects outliers by randomly constructing binary trees. Its core idea is that outliers are more likely to be isolated because they are different from most data points. , calculated by the following formula:

[0027] in: Each sample point Length of path in forest is its depth in the random tree; For sample The average path length in the random forest. The shorter the path length, the easier it is for the sample to be isolated, which means that it is more likely to be an anomaly. is the average path length of the binary tree, which depends on the number of samples , the calculation formula is: in For the The number of harmonics is approximately In(n)+0.577215 (Euler constant). This parameter is used to adjust the path length under different scale data sets.

[0028] Output of the Data Collection and Preprocessing Module: Anomaly Score The closer it is to 1, the more abnormal the sample is.

[0029] In addition, the data collection and preprocessing module of the present application includes a sliding window processing function for time series data, through which the data continuously collected by the sensor is processed in segments to generate a data set suitable for association rule mining.

[0030] The anomaly detection module of the present application determines which data points in the production data detected by the isolation forest algorithm are anomalies through a set threshold. The threshold is set according to business needs and data characteristics to distinguish between normal points and anomalies. This threshold can be set based on experience or optimized through cross-validation, grid search and other methods.

[0031] In this application, the abnormal classification module uses support vector machine (SVM) to classify the detected abnormal points. The classification fault types include equipment failure, environmental abnormality and operation error. SVM is a supervised classification algorithm that distinguishes different types of abnormal points by finding the hyperplane that maximizes the classification interval. The classification decision function of SVM is:

[0032] in: is the weight vector, indicating the direction of the hyperplane; is the bias term, which controls the position of the hyperplane; The symbol decision The category it belongs to (positive or negative).

[0033] By optimizing and , SVM finds a hyperplane that can maximize the class interval and classifies different types of anomalies (such as equipment failure, environmental anomalies, operational errors, etc.).

[0034] In this application, the root cause analysis module analyzes the possible causes of failures based on association rule mining technology, and uses support, confidence and lift to measure the effectiveness of the rules.

[0035] Association rule mining is used to mine factors that may cause a certain type of failure from historical data and generate rules, such as "If the temperature is > 100°C and the pressure is > 5MPa, it may cause equipment shutdown."

[0036] The root cause analysis module uses association rule mining technology to analyze the cause of the fault based on the following indicators: Support: indicates the frequency of event A and event B occurring simultaneously. The calculation formula is:

[0037] It indicates the proportion of events A and B occurring simultaneously in all data; the frequency of event B occurring, calculated as: , It indicates the proportion of event B in all data; Confidence: It indicates the probability of event B occurring when event A occurs. The calculation formula is: , It represents the probability of event B occurring when event A occurs. Lift: Indicates the strength of the association between event A and event B. The calculation formula is:

[0038] If Lift>1, it means there is a positive correlation between A and B; if Lift<1, it means the relationship between the two is weak, or even negatively correlated.

[0039] Application scenario: Through support, confidence and lift, the system can mine the cause of failure from historical data. For example, the association between "overtemperature" and "cooling system failure" can be clearly identified through association rules.

[0040] It should be noted that root cause analysis often involves not only a single level of fault causes, but also multi-level causal chains. In order to more accurately locate the deep-level root causes of faults, the system adopts a hierarchical association rule mining method, that is, the root cause analysis module uses hierarchical association rule mining technology, including: The first level of causal analysis identifies the direct causes of the failure; The second layer of causal analysis is to explore the potential causes of the first layer failures; The third and deeper causal analysis further analyzes the deep fault causes caused by equipment maintenance, operating behavior or other external factors Specifically, the first level of causal analysis: Identify the main causes of failures through association rule mining. For example: Rule: "Temperature > 100°C and device load > 80%" → "Bad soldering". Support: 0.3 (30% of the samples have both temperature and load exceeding the threshold).

[0041] Confidence level: 0.8 (80% of the samples will have poor welding when both temperature and load exceed the standard).

[0042] Second level causal analysis: Continue to analyze the potential causes of the first level failure. For example: Rule: "Cooling system failure" → "Temperature rise".

[0043] Support: 0.2, Confidence: 0.7, Lift: 1.5. This indicates that the failure of the cooling system directly caused the temperature to rise.

[0044] The third level of causal analysis: further analyze the root cause of the cooling system failure. For example: Rule: "Insufficient maintenance" → "Cooling system failure", with support and confidence of 0.15 and 0.85 respectively, indicates the deep-seated cause of the cooling system problem.

[0045] Hierarchical analysis summary: The system identifies the underlying root cause of a fault by gradually digging into the cause-and-effect chain. For example, insufficient equipment maintenance was ultimately identified as the root cause of cooling system failure and overheating.

[0046] It should be noted that in hierarchical association rule mining, although the same method (such as Apriori algorithm or FP-growth algorithm) can be used for rule mining at different levels, there are some differences in rule mining at each level, which are mainly reflected in the construction of input data sets and mining targets. Therefore, although the basic method is the same, it will be adjusted according to the level in actual application.

[0047] Specifically: 1. First-level association rule mining Goal: To discover direct anomaly correlation or the direct cause of a failure.

[0048] Dataset: The first level uses the original production data, including sensor data, equipment status and environmental data, etc. The mined rules reflect the factors that directly affect the system.

[0049] Mining process: Common association rule mining algorithms (such as the Apriori algorithm) mainly discover frequent condition items and result items through support, confidence, and lift.

[0050] Example: Too high temperature + too high equipment load → bad welding.

[0051] The exploration is of superficial cause-and-effect relationships, usually focusing on the direct impact of equipment and environmental anomalies on product quality.

[0052] 2. Second level association rule mining Objective: To further explore the deep-seated causes that lead to the first-level failure, that is, the deeper causes behind the first-level causes.

[0053] Dataset: The input data of the second level is based on the output of the first level, that is, the fault causes that have been found (such as high temperature) are used as conditions to be analyzed, and the factors affecting these conditions are further mined in combination with the original data.

[0054] Mining process: Similar to the first layer, but the focus of the data set has changed. Now the goal is to find factors that lead to problems such as "too high temperature".

[0055] Example: Cooling system malfunction → temperature too high.

[0056] The cause of the association here is no longer the production result, but the underlying equipment status.

[0057] 3. Mining association rules at the third and subsequent levels Goal: Continue to drill down to explore deeper causes of failures such as equipment, environment, or maintenance actions.

[0058] Dataset: At this point, the dataset will be combined with the original data based on the output of the second layer (such as cooling system failure). The goal is to find the reasons behind the cooling system failure, such as equipment maintenance issues or aging parts.

[0059] Mining process: As in the first two layers, the association rule mining algorithm is continued, but the goal at each layer is to trace the root cause of the failure in the upper layer.

[0060] Example: Inadequate maintenance → cooling system failure.

[0061] Ultimately this may reveal root causes such as inadequate equipment maintenance or process management issues.

[0062] Other optional adjustments: Adjustment of support and confidence thresholds: In deeper mining, you may need to lower the support threshold because deeper root causes usually occur less frequently, but they may be critical factors.

[0063] Selection of data features: As the level goes deeper, it may be necessary to reduce the dimension or transform some of the original data features and select data features that are more relevant to the current analysis level.

[0064] The adaptive learning module in this application updates the classification model and association rules based on the changes in real-time data, and uses the online learning mechanism to dynamically adjust the classification boundaries of the support vector machine to ensure that the classification model adapts to changes in the production environment. Specifically, it includes: Online learning mechanism: In order to cope with changes in the industrial environment, the system integrates an online learning module to continuously update the classification model and association rules based on new data. For example, through adaptive SVM, the system can dynamically adjust the hyperplane position when real-time data is updated, ensuring that the classification model always adapts to the latest production environment.

[0065] User feedback mechanism: The system allows operators to provide feedback on the analysis results. If the root cause analysis results are wrong or incomplete, users can make corrections. The system performs incremental learning based on these feedbacks to improve the accuracy of root cause analysis.

[0066] Those skilled in the art should understand that the embodiments of the present invention shown in the above description are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A system for equipment failure type classification and root cause analysis based on isolation forest algorithm, characterized by: The system comprises: The data collection and preprocessing module is used to collect environmental data, equipment status data and production result data from equipment sensors, and clean, process and standardize them. The outliers are detected by the isolation forest algorithm. Anomaly detection module, which uses a set threshold to determine anomalies in the data detected by the isolation forest algorithm, wherein the isolation forest algorithm detects anomalies that are different from data points by randomly constructing a binary tree; The abnormality classification module uses support vector machines to classify the detected abnormal points. The fault types include equipment failure, environmental abnormality and operation error. The root cause analysis module analyzes the root causes of failures based on association rule mining technology and uses support, confidence and lift to measure the effectiveness of the rules.

2. According to claim 1, a system for equipment failure type classification and root cause analysis based on isolation forest algorithm is characterized by: The anomaly score of the isolation forest algorithm is calculated by the following formula: in: Each sample point The length of the path in the forest, is its depth in the random tree; For sample Average path length in random forests; is the average path length of the binary tree, which depends on the number of samples and is calculated as follows: in For the The number of harmonics.

3. The equipment failure type classification and root cause analysis system based on the isolation forest algorithm according to claim 1 is characterized by: The classification decision function of the support vector machine is: in: is the weight vector; is the bias term, The calculation result is positive or negative. The category it belongs to is positive or negative.

4. The equipment failure type classification and root cause analysis system based on the isolation forest algorithm according to claim 1 is characterized by: The root cause analysis module uses association rule mining technology to analyze the cause of the fault according to the following indicators: Support: indicates the frequency of event A and event B occurring simultaneously. The calculation formula is: in, It represents the proportion of events A and B occurring simultaneously in all data; The frequency of event B is calculated as: , It indicates the proportion of event B in all data; Confidence: It indicates the probability of event B occurring when event A occurs. The calculation formula is: in, The probability of event B occurring when event A occurs; Lift: Indicates the strength of the association between event A and event B. The calculation formula is: in, is the strength of the association between event A and event B.

5. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 4 is characterized by: The root cause analysis module adopts hierarchical association rule mining technology, including: The first level of causal analysis identifies the direct causes of the failure; The second layer of causal analysis is to explore the potential causes of the first layer failures; The third and deeper causal analysis further analyzes the deep causes of failures caused by equipment maintenance, operating behavior or other external factors.

6. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 5 is characterized by: The hierarchical association rule mining technology uses different data sets at different levels. The input data of the first level is the original production data, including sensor data and equipment status data, and the input data of subsequent levels is a combination of the association results output by the previous level and the original production data.

7. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 6 is characterized by: As the level goes deeper, the support and confidence thresholds are adjusted dynamically. The deeper the level, the lower the support threshold is, to ensure the excavation of deep fault root causes.

8. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 1 is characterized by: The system includes an adaptive learning module that updates the classification model and association rules based on changes in real-time data, and uses an online learning mechanism to dynamically adjust the classification boundaries of the support vector machine to ensure that the classification model adapts to changes in the production environment.

9. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 8 is characterized by: The adaptive learning module receives user feedback on the analysis results through a user feedback mechanism, and the feedback results are used to adjust the association rule model of the root cause analysis module to improve the root cause analysis accuracy of the system.

10. The equipment failure type classification and root cause analysis system based on isolation forest algorithm according to claim 1, characterized in that: The data collection and preprocessing module also includes a time series data discretization step, which uses a sliding window method to discretize the time series data. Through this function, the data continuously collected by the sensor is processed in segments to generate a data set suitable for association rule mining.

Citation Information

Patent Citations

  • Motorcycle electrical system fault detection system

    CN117849512A

  • Stainless steel tube defect identification system based on machine learning

    CN118823472A

  • Abnormal behavior detection method based on bastion host

    CN118869365A

Cited By

  • Software fault multi-dimensional association relationship analysis method based on association rule mining

    CN119718942A

  • Fault detection early warning method and system of navigation mark data acquisition system, and medium

    CN120822126A

  • Communication machine room environment adaptive adjustment management system based on machine learning

    CN121115961A

  • Equipment dynamic control method and device based on edge cloud collaboration

    CN122172644A