A machine learning-based prediction system for the health status of industrial equipment

Abstract

The present invention relates to the technical field of industrial equipment health prediction, and provides an industrial equipment health status prediction system based on machine learning, wherein a data acquisition module acquires real-time data and historical data of industrial equipment, a data processing module performs feature extraction on the data, a fault chain analysis module analyzes the chain relationship between equipment failures based on the data processing module, and constructs a chain failure propagation model; the machine learning module uses the chain failure propagation model established by the data processing module and the fault chain analysis module as input features to establish an industrial equipment health status prediction model; the prediction module receives the equipment operation data transmitted in real time by the data processing module, and predicts the current health status of the equipment, and the early warning module issues an early warning according to the prediction result of the prediction module; the present invention mines the fault propagation information in the equipment operation data through the fault chain analysis module, discovers the potential chain failure risk in advance, and provides timely and accurate early warning information for manufacturers.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment health prediction, and in particular to an industrial equipment health status prediction system based on machine learning. Background Art

[0002] In the field of industrial production, the stable operation of industrial equipment is crucial to ensuring production efficiency, product quality and personnel safety. With the continuous development of science and technology, industrial equipment has become increasingly complex and diversified, and its functions and performance have been greatly improved, but at the same time it has also brought higher maintenance and management difficulties.

[0003] With the rapid development of information technology and sensor technology, a large number of sensors are installed on industrial equipment. These sensors can collect various data in real time during the operation of the equipment, such as temperature, pressure, vibration, flow, etc. These data contain rich information on the health status of the equipment, making it possible to achieve more accurate equipment maintenance.

[0004] Machine learning technology has powerful capabilities in data processing and analysis. It can mine hidden patterns and rules from a large amount of historical data and real-time data, thus providing strong support for the prediction of the health status of industrial equipment. By analyzing the data collected by sensors, machine learning algorithms can establish a mapping relationship between the health status of equipment and data features. For example, by analyzing the spectral characteristics of vibration data, it can be determined whether the mechanical parts of the equipment have problems such as wear and imbalance; by monitoring the changing trend of temperature data, it can be predicted whether the equipment has the risk of overheating.

[0005] The prior art still has the following deficiencies:

[0006] 1. There is a lack of prediction of potential chain failures after a failure or after a failure is repaired. Modern industrial equipment is often an interrelated complex system. For example, in chemical production, a temperature sensor failure in a reactor may cause abnormal temperature control. Even if this problem is discovered and corrected in time, it may cause other failures, such as excessive pressure caused by abnormal temperature control, causing pipeline rupture, chemical leakage and other chain failures, which ultimately affect the health of industrial equipment;

[0007] 2. The prediction of cascading failures is not timely enough. For example, in the power system of industrial equipment, the failure of a key transformer may cause voltage fluctuations in the power grid, which in turn may lead to hidden cascading failures in a series of equipment that relies on stable voltage to operate. Most existing technologies predict the failure of this transformer and then repair or prevent it. However, after the operation, they rarely consider the failure of the motor, electronic controller, etc. caused by the failure of the transformer or the change of parameters before the failure. If this impact is ignored and the operation continues, and then predictions are made based on changes in time series data, the health status of the industrial equipment may not be predicted in a timely manner, thereby affecting the normal operation of the entire production process or power supply area. Summary of the invention

[0008] In order to solve the above technical problems, the present invention provides an industrial equipment health status prediction system based on machine learning. By calculating the fault frequency, screening frequent item sets, establishing association rules to construct a fault propagation model and integrating equipment operating parameters as input features of the machine learning model, machine learning can predict the chain relationship between faults to solve the problems in the prior art.

[0009] A machine learning-based industrial equipment health status prediction system, comprising:

[0010] A data acquisition module, which is used to collect real-time data of industrial equipment from various sensors and monitoring systems of industrial equipment, and collect historical data from a historical fault database. The real-time data includes real-time parameter data of industrial equipment during operation and basic information data of industrial equipment. The historical data includes the time and type data of past faults of industrial equipment and historical operation parameter data;

[0011] A data processing module, which is data-connected to the data acquisition module and is used to extract features from the real-time operating parameter data and historical operating parameter data collected by the data acquisition module, wherein the extracted features include but are not limited to temperature features, vibration features, pressure features, current features, and wear degree features;

[0012] A fault chain analysis module, which is data-connected to the data processing module, analyzes the chain relationship between equipment failures based on the features extracted from historical failure data by the data processing module, and constructs a chain failure propagation model;

[0013] A machine learning module, wherein the machine learning module is data-connected with the data processing module and the fault chain analysis module, and constructs an industrial equipment health status prediction model through machine learning based on the data processed by the data processing module and the chain failure propagation model established by the fault chain analysis module as input features;

[0014] A prediction module, which is data-connected to the data processing module and the machine learning module, receives the equipment operation data transmitted in real time by the data processing module, inputs the feature vector obtained by real-time processing by the data processing module into the prediction model transmitted by the machine learning module, and predicts the current health status of the equipment, including whether there is a fault, the type of fault, the probability of the fault occurring, and the chain failure information caused;

[0015] The early warning module is connected to the prediction module data. According to the prediction results of the prediction module, the health status of the equipment or the existence of potential major risks is judged. If so, an early warning message is issued in time. The early warning message includes the equipment name, fault prediction details, and the consequences of the chain failure caused.

[0016] Preferably, the fault chain analysis module includes a chain rule mining unit, a fault propagation model building unit, an operation impact analysis unit and a data integration unit;

[0017] The chain rule mining unit calculates the frequency of each fault type, obtains its support by the ratio of the number of occurrences of the fault type to the total number of fault records, finds out the fault type combinations whose support is higher than a set threshold, i.e., frequent item sets, generates association rules representing the potential causal relationship between faults based on the frequent item sets, calculates the confidence of these association rules, and retains only the association rules whose confidence exceeds the specified value, which are the chain trigger relationships between the fault types;

[0018] The fault propagation model building unit uses a directed graph to build the fault propagation model based on the association rules calculated by the chain rule mining unit, and then uses an adjacency matrix to quantify the fault propagation model, and uses the adjacency matrix to show the fault propagation path and probability;

[0019] The operation impact analysis unit calculates the probability of the fault propagating to other faults based on each fault type and the values ​​of the operation parameters of the relevant equipment, and calculates it by the ratio of the number of co-occurrence and single occurrence of faults within a specific parameter value range;

[0020] The data integration unit integrates and optimizes the fault propagation model based on the fault propagation model adjacency matrix and the calculation results of the operation impact analysis unit to form a new adjacency matrix.

[0021] Preferably, the specific process of the chain rule mining unit mining the association rules between faults is as follows:

[0022] First, calculate the frequency of occurrence of a single fault. Suppose the number of times fault type T appears in the data set is count(T), then its support sup(T) is:

[0023]

[0024] Among them, n is the total number of fault records; then find the frequent item set, that is, the combination of fault types whose support is greater than the set threshold sup1. The support reflects the frequency of the fault type T in the data set and is an important indicator for judging whether the fault is universal.

[0025] For two fault types T1 and T2, the support of their combination is:

[0026]

[0027] If sup(T a ,T b )>sup1, then (T a ,T b ) is a frequent two-item set, where a and b are specific fault type indexes. By calculation, find out the fault type combinations that appear together in the data set. These combinations indicate that there is a potential correlation between the faults. Then, generate association rules based on the frequent item sets:

[0028]

[0029] Then calculate the confidence of each association rule

[0030]

[0031] Finally, the association rules with confidence greater than the set threshold con1 are retained. These rules represent the possible chain relationship between faults. The setting of threshold con1 and threshold sup1 can be manually set or obtained through grid search based on statistical analysis of historical fault data records.

[0032] Preferably, the specific construction and quantification process of the fault propagation model construction unit is as follows;

[0033] The fault propagation model is represented by a directed graph G=(T, E), where the node T represents the fault type and the edge E represents the relationship between each fault type, i.e., the association rule;

[0034] If there is an association rule Then there is a line from node T in the graph a Point to node T b The weight of the directed edge is set to the confidence of the association rule

[0035] At the same time, the adjacency matrix is ​​introduced to quantify the fault propagation model. Let C be the adjacency matrix of the fault propagation model. If there is a fault type T a To fault type T b The propagation relationship, that is, the directed edge, then otherwise

[0036] The elements in the adjacency matrix Non-zero, indicating that the fault type T a To fault type T b There is a potential propagation path, and the weight of the edge is the confidence of the association rule is the fault propagation probability.

[0037] Preferably, the specific analysis process of the operation impact analysis unit is as follows:

[0038] For a certain industrial equipment, suppose its operating parameter set V(V1, V2, V3, ... V u ) is a set of operating parameters of the industrial equipment that changes with time series, where u represents the index of the operating parameter;

[0039] Divide the equipment operation parameter data according to before and after the fault occurs, and calculate the probability of the fault propagating to a specific subsequent fault type under different parameter values ​​before and after the fault occurs;

[0040] Let P(T b |T a ,V u ) indicates that in the fault T a Occurs and the device operating parameter value is V u When the fault propagates to T b The probability of is obtained by statistically analyzing the failure data:

[0041]

[0042] Where [V1, V2] is a value interval of parameter V, V1 represents before the fault occurs, and V2 represents after the fault occurs; count(T a ∧T b ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the range of parameter v a Occurs and propagates to fault T b The number of times, count(T a ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the parameter value range. a The number of times it occurred.

[0043] Preferably, the data integration unit calculates the probability P(T b |T a,V), perform the following integration operations;

[0044] Let V represent the set of equipment operating parameters associated with fault type T. For each V u , we optimize as follows

[0045]

[0046] Among them, w(V u ) is the equipment operating parameter V u The weight of the parameter represents the relative importance of the parameter on fault propagation. The weight w is determined by calculating the parameter V u Fault Propagation The Pearson correlation coefficient was determined;

[0047] The optimized adjacency matrix The influence of equipment operating parameters on fault propagation is integrated into the fault propagation model G.

[0048] Preferably, the machine learning module includes a data optimization unit and a model training unit;

[0049] The data optimization unit extracts the basic features of the historical data from the data processing module to construct a basic feature matrix X, and then converts the node sequence in the fault propagation model directed graph G constructed in the fault chain analysis module, that is, the fault propagation path, into a feature vector, and encodes each fault propagation path by a one-hot encoding method to add it to the basic feature matrix X as a new feature vector. At the same time, the optimized fault propagation probability Added to the feature matrix X as additional probability features;

[0050] The model training unit learns and trains the industrial health status prediction model based on the feature matrix optimized by the data optimization unit through a neural network deep algorithm, and defines the predicted fault type problem in combination with a cross entropy loss function, wherein the feature matrix is ​​the core input data of the machine learning model.

[0051] Preferably, the data optimization unit specifically processes as follows:

[0052] Suppose the set of basic feature matrices extracted by the data processing module is X(X1, X2, X3, ... X0), where i represents a specific industrial equipment, and each feature matrix represents the matrix form of the operating parameter data of the industrial equipment i extracted at a specified time;

[0053] Next, suppose that industrial equipment i has a total of m types of faults. For each fault type T in the directed graph node sequence, create a vector of length m. If the fault type is Ta , then the ath position of the vector is 1, and the other positions are 0; connect the encoding vectors of all fault types in the path in sequence to form a fault propagation path feature vector and add it to the feature matrix X, and then add the optimized fault propagation probability As additional probabilistic features, each feature matrix X0 eventually becomes X0′.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention uses a chain rule mining unit to calculate support and confidence based on the frequency of fault occurrence to find association rules. The fault propagation model construction unit constructs a directed graph and a quantized adjacency matrix based on these rules to display the fault propagation path and probability. The operation impact analysis unit further considers the impact of equipment operation parameters on the probability of fault propagation and optimizes the model. The data integration unit integrates the relevant information and inputs it into the machine learning module. This series of processes enables the system to analyze the chain relationship between equipment failures more comprehensively and meticulously. Before a failure occurs or after a failure is repaired, the potential chain failure hazards can be predicted in advance to a certain extent based on the constructed chain failure propagation model. This can better maintain the overall health of industrial equipment.

[0056] 2. The present invention fully mines the hidden information in the equipment operation data through the fault chain analysis module and the machine learning module, establishes accurate fault propagation model and prediction model, realizes real-time monitoring and accurate prediction of the equipment health status, discovers potential fault hazards and chain failure risks in advance, and provides manufacturers with timely and accurate early warning information, so that they can take targeted preventive measures and optimize maintenance strategies, effectively reduce equipment failure rate, reduce unplanned downtime, improve production efficiency, and ensure product quality and production safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0058] Figure 2 It is a schematic diagram of the work flow of the fault chain analysis module of the present invention;

[0059] Figure 3 It is a schematic diagram of the workflow of the chain rule mining unit of the present invention. DETAILED DESCRIPTION

[0060] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0061] The present invention provides an industrial equipment health status prediction system based on machine learning, comprising:

[0062] A data acquisition module, which is used to collect real-time data of industrial equipment from various sensors and monitoring systems of industrial equipment, and collect historical data from a historical fault database. The real-time data includes real-time parameter data of industrial equipment during operation and basic information data of industrial equipment. The historical data includes the time and type data of past faults of industrial equipment and historical operation parameter data;

[0063] A data processing module, which is data-connected to the data acquisition module and is used to extract features from the real-time operating parameter data and historical operating parameter data collected by the data acquisition module, wherein the extracted features include but are not limited to temperature features, vibration features, pressure features, current features, and wear degree features;

[0064] A fault chain analysis module, which is data-connected to the data processing module, analyzes the chain relationship between equipment failures based on the features extracted from historical failure data by the data processing module, and constructs a chain failure propagation model;

[0065] A machine learning module, wherein the machine learning module is data-connected with the data processing module and the fault chain analysis module, and constructs an industrial equipment health status prediction model through machine learning based on the data processed by the data processing module and the chain failure propagation model established by the fault chain analysis module as input features;

[0066] A prediction module, which is data-connected to the data processing module and the machine learning module, receives the equipment operation data transmitted in real time by the data processing module, inputs the feature vector obtained by real-time processing by the data processing module into the prediction model transmitted by the machine learning module, and predicts the current health status of the equipment, including whether there is a fault, the type of fault, the probability of the fault occurring, and the chain failure information caused;

[0067] The early warning module is connected to the prediction module data. According to the prediction results of the prediction module, the health status of the equipment or the existence of potential major risks is judged. If so, an early warning message is issued in time. The early warning message includes the equipment name, fault prediction details, and the consequences of the chain failure caused.

[0068] Example:

[0069] like Figure 1-3As shown, in this embodiment, in a large industrial production plant, there is a complex industrial equipment system, which is composed of multiple sub-equipment, including heating furnaces, rolling mills, cooling devices, pumps, compressors, pipeline systems and many other closely related precision equipment. The sub-equipment is interconnected and works together. Failure in any link may trigger a chain reaction, affecting the continuity of the entire production process and product quality, and may even cause equipment damage, personnel safety accidents and huge economic losses.

[0070] Therefore, this system is introduced to timely predict the propagation risks when a fault occurs, and to predict the health status of industrial equipment when a fault has occurred or is being repaired, to find out the hidden fault propagation risks;

[0071] The data acquisition module is installed on relevant industrial equipment, including temperature sensors, pressure sensors, liquid level sensors, vibration sensors, and current sensors, to obtain the operating status information of the equipment;

[0072] At the same time, historical data such as equipment failure time, failure type, and operating parameters at the time are extracted from the factory's historical maintenance records and fault reports. Basic information about the equipment, such as equipment model, age, manufacturer, etc., is also collected and summarized and transmitted to the data processing module.

[0073] The data processing module extracts features from the collected real-time operating parameter data and historical data, such as heating rate, cooling rate, temperature fluctuation range, and vibration features such as equipment vibration frequency, amplitude, and vibration acceleration.

[0074] The fault chain analysis module includes a chain rule mining unit, a fault propagation model building unit, an operation impact analysis unit, and a data integration unit;

[0075] The chain rule mining unit finds out the frequently occurring fault combinations by mining the historical fault data. First, the frequency of each fault type is calculated, and its support is obtained by the ratio of the number of occurrences of the fault type to the total number of fault records. The fault type combinations with support higher than the set threshold, i.e., frequent item sets, are found. Based on the frequent item sets, association rules representing the potential causal relationship between faults are generated, and the confidence of these association rules is calculated. Only the association rules with confidence exceeding the specified value are retained. These rules are the chain trigger relationships between the fault types.

[0076] The fault propagation model building unit uses a directed graph to build a fault propagation model based on the association rules calculated by the chain rule mining unit, and then uses the adjacency matrix to quantify the fault propagation model and display the fault propagation path and probability.

[0077] The operation impact analysis unit calculates the probability of the fault propagating to other faults based on each fault type and the values ​​of the operation parameters of the relevant equipment, and calculates it by the ratio of the number of co-occurrence and single occurrence of faults within a specific parameter value range;

[0078] The data integration unit integrates and optimizes the fault propagation model based on the fault propagation model adjacency matrix and the calculation results of the operation impact analysis unit to form a new adjacency matrix.

[0079] The specific process of mining association rules between faults by the chain rule mining unit is as follows:

[0080] First, calculate the frequency of occurrence of a single fault. Suppose the number of times fault type T appears in the data set is count(T), then its support sup(T) is:

[0081]

[0082] Among them, n is the total number of fault records; then find the frequent item set, that is, the combination of fault types whose support is greater than the set threshold sup1. The support reflects the frequency of the fault type T in the data set and is an important indicator for judging whether the fault is universal.

[0083] For two fault types T1 and T2, the support of their combination is:

[0084]

[0085] If sup(T a ,T b )>sup1, then (T a ,T b ) is a frequent two-item set, where a and b are specific fault type indexes. By calculation, find out the fault type combinations that appear together in the data set. These combinations indicate that there is a potential correlation between the faults. Then, generate association rules based on the frequent item sets:

[0086]

[0087] Then calculate the confidence of each association rule

[0088]

[0089] Finally, the association rules with confidence greater than the set threshold con1 are retained. These rules represent the possible chain relationship between faults. The setting of threshold con1 and threshold sup1 can be manually set or obtained through grid search based on statistical analysis of historical fault data records.

[0090] The specific construction and quantification process of the fault propagation model construction unit is as follows;

[0091] The fault propagation model is represented by a directed graph G = (T, E), where the node T represents the fault type and the edge E represents the relationship between each fault type, i.e., the association rule;

[0092] If there is an association rule Then there is a line from node T in the graph a Point to node T b The weight of the directed edge is set to the confidence of the association rule

[0093] At the same time, the adjacency matrix is ​​introduced to quantify the fault propagation model. Let C be the adjacency matrix of the fault propagation model. If there is a fault type T a To fault type T b The propagation relationship, that is, the directed edge, then otherwise

[0094] The elements in the adjacency matrix Non-zero, indicating that the fault type T a To fault type T b There is a potential propagation path, and the weight of the edge is the confidence of the association rule is the fault propagation probability.

[0095] The specific analysis process of the operation impact analysis unit is as follows:

[0096] For a certain industrial equipment, suppose its operating parameter set V(V1, V2, V3, ... V u ) is a set of operating parameters of the industrial equipment that changes with time series, where u represents the index of the operating parameter;

[0097] Divide the equipment operation parameter data according to before and after the fault occurs, and calculate the probability of the fault propagating to a specific subsequent fault type under different parameter values ​​before and after the fault occurs;

[0098] Let P(T b |T a ,V u ) indicates that in the fault T a Occurs and the device operating parameter value is V u When the fault propagates to T b The probability of is obtained by statistically analyzing the failure data:

[0099]

[0100] Where [V1, V2] is a value interval of parameter V, V1 represents before the fault occurs, and V2 represents after the fault occurs; count(T a ∧T b ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the range of parameter v a Occurs and propagates to fault T b The number of times, count(T a ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the parameter value range. a The number of times it occurred.

[0101] Preferably, the data integration unit calculates the probability P(T b |T a ,V), perform the following integration operations;

[0102] Let V represent the set of equipment operating parameters associated with fault type T. For each V u , we optimize as follows

[0103]

[0104] Among them, w(V u ) is the equipment operating parameter V u The weight of the parameter represents the relative importance of the parameter on fault propagation. The weight w is determined by calculating the parameter V u Fault Propagation The Pearson correlation coefficient was determined;

[0105] The optimized adjacency matrix The influence of equipment operating parameters on fault propagation is integrated into the fault propagation model G.

[0106] The machine learning module includes a data optimization unit and a model training unit;

[0107] The data optimization unit extracts the basic features of historical data from the data processing module to construct the basic feature matrix X, and then converts the node sequence in the fault propagation model directed graph G constructed in the fault chain analysis module, that is, the fault propagation path, into a feature vector, and encodes each fault propagation path through the one-hot encoding method to add it to the basic feature matrix X as a new feature vector. At the same time, the optimized fault propagation probability Added to the feature matrix X as additional probability features;

[0108] The data optimization unit specifically processes as follows:

[0109] Suppose the set of basic feature matrices extracted by the data processing module is X(X1, X2, X3, ... X0), where i represents a specific industrial equipment, and each feature matrix represents the matrix form of the operating parameter data of the industrial equipment i extracted at a specified time;

[0110] Next, suppose that industrial equipment i has a total of m types of faults. For each fault type T in the directed graph node sequence, create a vector of length m. If the fault type is T a , then the ath position of the vector is 1, and the other positions are 0; connect the encoding vectors of all fault types in the path in sequence to form a fault propagation path feature vector and add it to the feature matrix X, and then add the optimized fault propagation probability As additional probabilistic features, each feature matrix X0 eventually becomes X0′.

[0111] The model training unit is based on the feature matrix optimized by the data optimization unit. Through the neural network deep algorithm, the optimized feature matrix is ​​learned and trained to learn and train the industrial health status prediction model. A large amount of historical data and some real-time data are used as training sets. By adjusting the parameters and structure of the model, the model can accurately predict whether the equipment has a fault, the type of fault, the time when the fault occurs, and the possible chain failures.

[0112] During the training process, the cross entropy loss function is used to evaluate the prediction error of the model, and the parameters of the model are continuously optimized through the back propagation algorithm to improve the prediction accuracy of the model. Among them, the feature matrix is ​​the core input data of the machine learning model.

[0113] After the training is completed, the prediction module receives the real-time equipment operation data from the data processing module, converts it into a feature vector, and inputs it into the trained prediction model. For example, when the model receives real-time data features such as the temperature of a certain device is too high or the radiator power is too high, it can predict that the device may have a risk of failure, and the failure type is temperature out of control. At the same time, it is expected that subsequent chain failures may occur, such as radiator overload damage, etc., thereby providing early warning information to maintenance personnel.

[0114] The early warning module will issue early warning information in a timely manner when it determines that the equipment has health problems or potential major risks based on the prediction results of the prediction module. The early warning information includes the equipment name, fault prediction details, possible chain failure consequences, and recommended countermeasures. The early warning information will be sent to relevant equipment maintenance personnel, production management personnel, and safety supervisors through the factory's internal monitoring system, SMS notifications, sound and light alarms, and other methods;

[0115] In a highly complex and dangerous environment such as industrial production, the spread of equipment failures is often difficult to fully and accurately predict and prevent based on expert experience. Many failures in the production process do not exist in isolation, but are interconnected and influenced by complex physical or chemical processes. Their propagation paths and impact levels are restricted by a variety of factors. The interactions between these factors are highly nonlinear and uncertain, which makes it difficult and challenging for experts to judge the spread of failures.

[0116] This system uses the fault chain analysis module and machine learning module to fully explore the hidden information in the equipment operation data, establish accurate fault propagation models and prediction models, realize real-time monitoring and accurate prediction of the equipment health status, discover potential fault hazards and chain failure risks in advance, and provide manufacturers with timely and accurate early warning information, so that they can take targeted preventive measures and optimize maintenance strategies, effectively reduce equipment failure rates, reduce unplanned downtime, improve production efficiency, and ensure product quality and production safety.

[0117] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An industrial equipment health status prediction system based on machine learning, characterized in that: include: A data acquisition module, which is used to collect real-time data of industrial equipment from various sensors and monitoring systems of industrial equipment, and collect historical data from a historical fault database. The real-time data includes real-time parameter data of industrial equipment during operation and basic information data of industrial equipment. The historical data includes the time and type data of past faults of industrial equipment and historical operation parameter data; A data processing module, which is data-connected to the data acquisition module and is used to extract features from the real-time operating parameter data and historical operating parameter data collected by the data acquisition module, wherein the extracted features include but are not limited to temperature features, vibration features, pressure features, current features, and wear degree features; A fault chain analysis module, which is data-connected to the data processing module, analyzes the chain relationship between equipment failures based on the features extracted from historical failure data by the data processing module, and constructs a chain failure propagation model; A machine learning module, wherein the machine learning module is data-connected with the data processing module and the fault chain analysis module, and constructs an industrial equipment health status prediction model through machine learning based on the data processed by the data processing module and the chain failure propagation model established by the fault chain analysis module as input features; A prediction module, which is data-connected to the data processing module and the machine learning module, receives the equipment operation data transmitted in real time by the data processing module, inputs the feature vector obtained by real-time processing by the data processing module into the prediction model transmitted by the machine learning module, and predicts the current health status of the equipment, including whether there is a fault, the type of fault, the probability of the fault occurring, and the chain failure information caused; The early warning module is connected to the prediction module data. According to the prediction results of the prediction module, the health status of the equipment or the existence of potential major risks is judged. If so, an early warning message is issued in time. The early warning message includes the equipment name, fault prediction details, and the consequences of the chain failure caused.

2. The machine learning-based industrial equipment health status prediction system according to claim 1, characterized in that: The fault chain analysis module includes a chain rule mining unit, a fault propagation model building unit, an operation impact analysis unit and a data integration unit; The chain rule mining unit calculates the frequency of each fault type, obtains its support by the ratio of the number of occurrences of the fault type to the total number of fault records, finds out the fault type combinations whose support is higher than a set threshold, i.e., frequent item sets, generates association rules representing the potential causal relationship between faults based on the frequent item sets, calculates the confidence of these association rules, and retains only the association rules whose confidence exceeds the specified value, which are the chain trigger relationships between the fault types; The fault propagation model building unit uses a directed graph to build the fault propagation model based on the association rules calculated by the chain rule mining unit, and then uses an adjacency matrix to quantify the fault propagation model, and uses the adjacency matrix to show the fault propagation path and probability; The operation impact analysis unit calculates the probability of the fault propagating to other faults based on each fault type and the values ​​of the operation parameters of the relevant equipment, and calculates it by the ratio of the number of co-occurrence and single occurrence of faults within a specific parameter value range; The data integration unit integrates and optimizes the fault propagation model based on the fault propagation model adjacency matrix and the calculation results of the operation impact analysis unit to form a new adjacency matrix.

3. The machine learning-based industrial equipment health status prediction system according to claim 2, characterized in that: The specific process of mining association rules between faults by the chain rule mining unit is as follows: First, calculate the frequency of occurrence of a single fault. Suppose the number of times fault type T appears in the data set is count(T), then its support sup(T) is: Among them, n is the total number of fault records; then find the frequent item set, that is, the combination of fault types whose support is greater than the set threshold sup1. The support reflects the frequency of the fault type T in the data set and is an important indicator for judging whether the fault is universal. For two fault types T1 and T2, the support of their combination is: If sup(T a ,T b )>sup1, then (T a ,T b ) is a frequent two-item set, where a and b are specific fault type indexes. By calculation, find out the fault type combinations that appear together in the data set. These combinations indicate that there is a potential correlation between the faults. Then, generate association rules based on the frequent item sets: Then calculate the confidence of each association rule Finally, the association rules with confidence greater than the set threshold con1 are retained. These rules represent the possible chain relationship between faults. The setting of threshold con1 and threshold sup1 can be manually set or obtained through grid search based on statistical analysis of historical fault data records.

4. The industrial equipment health status prediction system based on machine learning as claimed in claim 2, characterized in that: The specific construction and quantification process of the fault propagation model construction unit is as follows: The fault propagation model is represented by a directed graph G=(T, E), where the node T represents the fault type and the edge E represents the relationship between each fault type, i.e., the association rule; If there is an association rule Then there is a line from node T in the graph a Point to node T b The weight of the directed edge is set to the confidence of the association rule At the same time, the adjacency matrix is ​​introduced to quantify the fault propagation model. Let C be the adjacency matrix of the fault propagation model. If there is a fault type T a To fault type T b The propagation relationship, that is, the directed edge, then otherwise The elements in the adjacency matrix Non-zero, indicating that the fault type T a To fault type T b There is a potential propagation path, and the weight of the edge is the confidence of the association rule is the fault propagation probability.

5. The industrial equipment health status prediction system based on machine learning as claimed in claim 2, characterized in that: The specific analysis process of the operation impact analysis unit is as follows: For a certain industrial equipment, suppose its operating parameter set V(V1, V2, V3, ... V u ) is a set of operating parameters of the industrial equipment that changes with time series, where u represents the index of the operating parameter; Divide the equipment operation parameter data according to before and after the fault occurs, and calculate the probability of the fault propagating to a specific subsequent fault type under different parameter values ​​before and after the fault occurs; Let P(T b |T a ,V u ) indicates that in the fault T a Occurs and the device operating parameter value is V u When the fault propagates to T b The probability of is obtained by statistically analyzing the failure data: Where [V1, V2] is a value interval of parameter V, V1 represents before the fault occurs, and V2 represents after the fault occurs; count(T a ∧T b ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the range of parameter v a Occurs and propagates to fault T b The number of times, count(T a ∧V u (V u ∈[V1,V2])) indicates that the fault T is within the parameter value range. a The number of times it occurred.

6. The industrial equipment health status prediction system based on machine learning as claimed in claim 2, characterized in that: The data integration unit calculates the probability P(T) of the influence of equipment operation parameters on fault propagation based on the adjacency matrix C of association rules. b |T a ,V), perform the following integration operations; Let V represent the set of equipment operating parameters associated with fault type T. For each V u , we optimize as follows Among them, w(V u ) is the equipment operating parameter V u The weight of the parameter represents the relative importance of the parameter on fault propagation. The weight w is determined by calculating the parameter V u Fault Propagation The Pearson correlation coefficient was determined; The optimized adjacency matrix The influence of equipment operating parameters on fault propagation is integrated into the fault propagation model G.

7. The industrial equipment health status prediction system based on machine learning as claimed in claim 6, characterized in that: The machine learning module includes a data optimization unit and a model training unit; The data optimization unit extracts the basic features of the historical data from the data processing module to construct a basic feature matrix X, and then converts the node sequence in the fault propagation model directed graph G constructed in the fault chain analysis module, that is, the fault propagation path, into a feature vector, and encodes each fault propagation path by a one-hot encoding method to add it to the basic feature matrix X as a new feature vector. At the same time, the optimized fault propagation probability Added to the feature matrix X as additional probability features; The model training unit learns and trains the industrial health status prediction model based on the feature matrix optimized by the data optimization unit through a neural network deep algorithm, and defines the predicted fault type problem in combination with a cross entropy loss function, wherein the feature matrix is ​​the core input data of the machine learning model.

8. The machine learning-based industrial equipment health status prediction system according to claim 7, characterized in that: The data optimization unit specifically processes as follows: Suppose the set of basic feature matrices extracted by the data processing module is X(X1, X2, X3, ... X0), where i represents a specific industrial equipment, and each feature matrix represents the matrix form of the operating parameter data of the industrial equipment i extracted at a specified time; Next, suppose that industrial equipment i has a total of m types of faults. For each fault type T in the directed graph node sequence, create a vector of length m. If the fault type is T a , then the ath position of the vector is 1, and the other positions are 0; connect the encoding vectors of all fault types in the path in sequence to form a fault propagation path feature vector and add it to the feature matrix X, and then add the optimized fault propagation probability As additional probabilistic features, each feature matrix X0 eventually becomes X0′.

Images