A method for analyzing equipment failure based on energy consumption big data

Abstract

A method for analyzing equipment failures based on energy consumption big data proposes to record and learn energy consumption data using association rules and multivariate state estimation technology, establish an evaluation model for equipment status, and determine the safe range of equipment operating parameters, so as to detect abnormal behavior or potential failures of equipment and provide users with feedback on equipment energy consumption, operating status, and equipment failure probability. The present invention achieves a deep integration of the online energy consumption monitoring system and the equipment failure early warning system, establishes an efficient and concise monitoring system, and realizes energy consumption monitoring while providing early warning of failures.

Description

Technical Field

[0001] The present invention belongs to the field of online monitoring of energy consumption in smart factories. It analyzes energy consumption data through multivariate state estimation technology and association rules, mines equipment failure characteristics from historical data, establishes a fault classification model, monitors potential failures of equipment or systems, and predicts possible failures of equipment. Background Art

[0002] The existing online energy consumption monitoring system is an intelligent system that monitors and manages the energy usage of manufacturing facilities in real time. The system can help companies optimize energy efficiency, reduce costs, and minimize environmental impact. The key components of the online energy consumption monitoring system are: sensor integration, data acquisition, data analysis and visualization, fault detection and diagnosis, energy efficiency recommendations, historical data storage, etc.

[0003] The equipment failure early warning system is a technology-driven solution designed to monitor and predict potential failures of industrial equipment, machinery or systems. The equipment failure early warning system usually uses sensors, data analysis and predictive maintenance technology to achieve its goals, and has a lot of overlap with the energy consumption online monitoring system in terms of system structure. At present, some companies have achieved interoperability and sharing of the two systems in terms of data transmission networks, data storage, and visualization panels, which has reduced the installation cost to a certain extent. However, these cost savings are negligible compared to the costs of sensors and monitoring and analysis systems that cannot be shared. Therefore, the energy consumption online monitoring system and the equipment failure early warning system are essentially two independently operated systems with difficult data integration. For enterprises, running these two independent monitoring systems not only faces great pressure in terms of installation costs, but also requires huge manpower costs in daily use and maintenance, resulting in a waste of resources. Summary of the invention

[0004] In order to overcome the shortcomings of the existing technology, to achieve the deep integration of the online energy consumption monitoring system and the equipment fault warning system, to establish an efficient and concise monitoring system, to realize energy consumption monitoring and fault warning at the same time, the present invention provides an equipment failure analysis method based on energy consumption big data. The present invention proposes to use association rules and multivariate state estimation technology to record and learn energy consumption data, establish an evaluation model for equipment status, and determine the safe range of equipment operating parameters, so as to detect abnormal behavior or potential failures of the equipment, and feedback information such as equipment energy consumption, operating status and equipment failure probability to users.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] A method for analyzing equipment failure based on energy consumption big data includes the following steps:

[0007] 1) Data collection: Collect and aggregate energy consumption data of devices, including power consumption levels, voltage fluctuations, and current usage information, using sensors and smart meters to capture real-time energy consumption data;

[0008] 2) Data preprocessing: Clean and preprocess the raw data to handle missing values, outliers, and inconsistencies; convert the data into a set format for analysis to ensure compatibility with the selected analysis method; extract relevant features from the energy consumption data to gain insight into device performance, including average power consumption, peak demand, and power peak frequency;

[0009] 3) Big Data Analysis: Apply association rules and multivariate state estimation techniques to learn equipment operation energy consumption data to identify patterns, trends, and anomalies in energy consumption data; classify normal and abnormal equipment behaviors, and use historical data to train models to identify patterns associated with various types of failures;

[0010] 4) Setting thresholds: Define safety thresholds for different parameters based on the energy consumption data when the equipment is working normally. When the actual data deviates significantly from the safety threshold, it may indicate a potential equipment failure.

[0011] 5) Alarm generation: When the analysis identifies a potential equipment failure, a notification or alarm is generated through the alarm system, which is sent to maintenance personnel or system operators for further investigation.

[0012] Furthermore, in the step 3), the processing process of the multivariate state estimation technology MSET is as follows: MSET is a nonlinear multivariate predictive diagnostic technology. By analyzing and comparing the actual monitoring parameters with the health data of the equipment during normal operation, various parameters during normal operation are calculated and estimated, and this normal state estimate is used as the standard; when the actual operating data is obtained, the health data is also used as the basis, and the degree of correlation between the actual data and the health data is found to estimate the actual operating state; the degree of correlation is determined by a weight vector, which is used to measure the similarity between the actual state and the normal state, and finally the estimated results of the health state and the actual operating state are compared and analyzed, and the concept of residual is introduced, and finally the equipment is diagnosed.

[0013] Furthermore, in the step 3), the association rule is as follows: Association rule is a basic concept in data mining and machine learning, especially in the field of exploratory data analysis, it is widely used to identify the relationship between variables or items in large data sets; Association rules usually take the form of "if-then" statements to express the relationship between different items in the data set. On the basis of multivariate state estimation technology, through Apriori association rules, by mining the association between energy consumption data when the equipment is in different fault states, the association between the most common faults in equipment failures and other faults is studied, and on this basis, early warnings are given for possible faults, and corresponding maintenance suggestions are given to maintenance personnel.

[0014] Preferably, the association rule is based on the existence of a certain regularity between the values ​​of two or more variables, which is called association. The association rule is expressed as an implication formula in the form of X→Y, where X and Y are respectively called the predecessor and successor of the association rule. The support, confidence and relevance of the association rule X→Y are used to measure the strength and reliability of the rule. Assume that there is a transaction data set P = {p1, p2, p3, ..., p k}, this transaction data set has k data, and the events represented by X and Y belong to this set;

[0015] The support S of the association rule X→Y is defined as the proportion of X∪Y tuples contained in the data set P, which can be expressed in the following form using probability theory:

[0016]

[0017] Represents the probability of item sets X and Y occurring simultaneously;

[0018] The confidence C of the association rule X→Y represents the probability that item set Y will occur if item set X occurs:

[0019]

[0020] The relevance L of the association rule X→Y is the ratio of the confidence C to the support S, which reflects how much the probability of the occurrence of X changes the occurrence of Y:

[0021]

[0022] Among them, the correlation Lift(X→Y) is the most important value in the association rule. It indicates the strength of the association between events X and Y, and represents the correlation between fault X and other faults Y. When Lift(X→Y)>1, it means that the occurrence of fault X will increase the occurrence of fault Y; when Lift(X→Y)<1, it means that the occurrence of fault X will inhibit the occurrence of fault Y.

[0023] Furthermore, the association rule mining process includes the following steps:

[0024] 3.1) Data preparation: This involves collecting and preprocessing the dataset to ensure that it is in a format suitable for analysis;

[0025] 3.2) Frequent Itemset Generation: A frequent itemset is a group of items (e.g., energy consumption data when a device has different types of failures) that appear together in a sufficient number of instances in the dataset. The Apriori algorithm is used to identify frequent itemsets.

[0026] 3.3) Association rule generation: Once the frequent itemsets are identified, association rules are generated from them. The association rules take the form of “if {X} then {Y}”, where X and Y are sets of items representing different types of equipment failures;

[0027] 3.4) Rule evaluation and pruning: In this step, the generated association rules are evaluated based on their interestingness measures (e.g., support, confidence, relevance) and may be pruned to remove rules that do not meet certain criteria or are not considered interesting;

[0028] 3.5) Interpretation and Visualization: The discovered association rules are interpreted and visualized to gain insight into the relationships between items in the dataset.

[0029] The beneficial effects of the present invention are mainly manifested in: achieving deep integration of the online energy consumption monitoring system and the equipment fault early warning system, establishing an efficient and concise monitoring system, and realizing energy consumption monitoring while providing early warning of faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a system architecture diagram for predicting equipment failures based on energy consumption big data.

[0031] Figure 2 It is a schematic diagram of the confidence and relevance of the item set after secondary screening. DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings.

[0033] Reference Figure 1 and Figure 2 , a method for analyzing equipment failure based on energy consumption big data, comprising the following steps:

[0034] 1) Data collection: Collect and aggregate energy consumption data of devices, including power consumption levels, voltage fluctuations, and current usage information, using sensors and smart meters to capture real-time energy consumption data;

[0035] 2) Data preprocessing: Clean and preprocess the raw data to handle missing values, outliers, and inconsistencies; convert the data into a set format for analysis to ensure compatibility with the selected analysis method; extract relevant features from the energy consumption data to gain insight into device performance, including average power consumption, peak demand, and power peak frequency;

[0036] 3) Big Data Analysis: Apply association rules and multivariate state estimation techniques to learn equipment operation energy consumption data to identify patterns, trends, and anomalies in energy consumption data; classify normal and abnormal equipment behaviors, and use historical data to train models to identify patterns associated with various types of failures;

[0037] In the step 3), the processing process of the multivariate state estimation technology MSET is as follows: MSET is a nonlinear multivariate predictive diagnostic technology. By analyzing and comparing the actual monitoring parameters with the health data of the equipment during normal operation, various parameters during normal operation are calculated and estimated, and this normal state estimate is used as the standard; when the actual operating data is obtained, the health data is also used as the basis, and the degree of correlation between the actual data and the health data is found to estimate the actual operating state; the degree of correlation is determined by a weight vector, which is used to measure the similarity between the actual state and the normal state. Finally, the estimated results of the health state and the actual operating state are compared and analyzed, and the concept of residual is introduced, and finally the equipment is diagnosed.

[0038] In the step 3), the association rule is as follows: association rule is a basic concept in data mining and machine learning, especially in the field of exploratory data analysis, it is widely used to identify the relationship between variables or items in large data sets; association rules usually take the form of "if-then" statements to represent the relationship between different items in the data set. On the basis of multivariate state estimation technology, through Apriori association rules, by mining the association between energy consumption data when the equipment is in different fault states, the association between the most common faults in equipment failures and other faults is studied, and on this basis, early warnings are given for possible faults, and corresponding maintenance suggestions are given to maintenance personnel.

[0039] Preferably, the association rule is based on the existence of a certain regularity between the values ​​of two or more variables, which is called association. The association rule is expressed as an implication formula in the form of X→Y, where X and Y are respectively called the predecessor and successor of the association rule. The support, confidence and relevance of the association rule X→Y are used to measure the strength and reliability of the rule. Assume that there is a transaction data set P = {p1, p2, p3, ..., p k}, this transaction data set has k data, and the events represented by X and Y belong to this set;

[0040] The support S of the association rule X→Y is defined as the proportion of X∪Y tuples contained in the data set P, which can be expressed in the following form using probability theory:

[0041]

[0042] Represents the probability of item sets X and Y occurring simultaneously;

[0043] The confidence C of the association rule X→Y represents the probability that item set Y will occur if item set X occurs:

[0044]

[0045] The relevance L of the association rule X→Y is the ratio of the confidence C to the support S, which reflects how much the probability of the occurrence of X changes the occurrence of Y:

[0046]

[0047] Among them, the correlation Lift(X→Y) is the most important value in the association rule. It indicates the strength of the association between events X and Y, and represents the correlation between fault X and other faults Y. When Lift(X→Y)>1, it means that the occurrence of fault X will increase the occurrence of fault Y; when Lift(X→Y)<1, it means that the occurrence of fault X will inhibit the occurrence of fault Y.

[0048] Furthermore, the association rule mining process includes the following steps:

[0049] 3.1) Data preparation: This involves collecting and preprocessing the dataset to ensure that it is in a format suitable for analysis;

[0050] 3.2) Frequent Itemset Generation: A frequent itemset is a group of items (e.g., energy consumption data when a device has different types of failures) that appear together in a sufficient number of instances in the dataset. The Apriori algorithm is used to identify frequent itemsets.

[0051] 3.3) Association rule generation: Once the frequent itemsets are identified, association rules are generated from them. The association rules take the form of “if {X} then {Y}”, where X and Y are sets of items representing different types of equipment failures;

[0052] 3.4) Rule evaluation and pruning: In this step, the generated association rules are evaluated based on their interestingness measures (e.g., support, confidence, relevance) and may be pruned to remove rules that do not meet certain criteria or are not considered interesting;

[0053] 3.5) Interpretation and Visualization: The discovered association rules are interpreted and visualized to gain insight into the relationships between items in the dataset.

[0054] 4) Setting thresholds: Define safety thresholds for different parameters based on the energy consumption data when the equipment is working normally. When the actual data deviates significantly from the safety threshold, it may indicate a potential equipment failure.

[0055] 5) Alarm generation: When the analysis identifies a potential equipment failure, a notification or alarm is generated through the alarm system, which is sent to maintenance personnel or system operators for further investigation.

[0056] In this embodiment, the specific implementation process of the multivariable state estimation technology (MSET) is as follows:

[0057] First, the observed n different energy consumption data are generated into an observation vector X(t i )=[x(t i )1,x(t i )2,x(t i )3,......,x(t i ) n ], where t i After observing the energy-consuming equipment for a long time, the energy consumption data matrix of the equipment can be obtained.

[0058]

[0059] Each row of the matrix is ​​a parameter from t1 to t m Each column of the matrix lists the time series values ​​from x1 to x n All parameters at a time t i So the matrix contains the values ​​of t1-t m Within a certain period of time, there are n different parameters of the equipment's energy consumption, such as power consumption, voltage, current, air pressure, and other energy consumption data. When the equipment is in normal operation, the healthy energy consumption data matrix D of the equipment can be obtained. H In the process of building a healthy data matrix, the computer completes the learning of the data characteristics of normal operation of the equipment and completes the construction of safety thresholds.

[0060] When the equipment begins to have potential failures, its observation data may deviate from the normal mode. At this time, if a new observation vector is input into the model based on historical data, the model may not be able to accurately estimate this new state. The direct result of the decrease in estimation accuracy is that the real-time observation vector X x With the estimated vector X y The difference between (i.e., residual error) will increase. Real-time observation vector X x and the estimated vector Xy The residual between

[0061] ε=X x -X y .

[0062] The real-time observation vector X x comes from real-time data observed by sensors, while X y =D H W, W is the weight matrix. In ||ε|| 2 Under the minimum constraint conditions, the least squares method can be used to obtain W, that is:

[0063]

[0064] represents a nonlinear operator. The most common nonlinear operators are the Euclidean norm, city block distance, linear correlation coefficient, and Gaussian operator. So the estimated vector is

[0065] The residual actually represents the difference between the observed value and the estimated value. In multivariate state estimation, a residual vector is calculated at each observation moment. By monitoring the size and change trend of the residual, it can be determined whether the equipment begins to be abnormal or malfunction. When the residual exceeds a certain preset safety threshold, it means that the equipment state is no longer normal, which helps operators and maintenance personnel to promptly discover and deal with potential problems, thereby improving the reliability and safety of the equipment.

[0066] The specific implementation process of association rule:

[0067] The Apriori principle is an algorithm for learning association rules. Its core idea is to find frequent item sets in the database through an iterative method of layer-by-layer search, thereby generating association rules. Under real conditions, there are many types of equipment failures, and the corresponding energy consumption data is also very large. If a traversal algorithm is used, it will be a huge challenge for computing time. The Apriori algorithm is based on an important concept: support S, confidence C, and correlation L. First, traverse the situation of 1 data combination, eliminate the data items with support lower than the minimum support, and then combine the remaining data to reduce the computing time. Traverse the situation of 2 data combinations, and then eliminate the combination with correlation L less than 1. Keep recursing until there are no more data combinations, which will greatly shorten the computing time.

[0068] First, a fault database is established. For example, if a certain device has 8 faults during operation, the abnormal energy consumption data corresponding to the 8 faults are represented by codes G1, G2, G3, G4, G5, G6, G7, and G8 respectively. The corresponding relationship between fault type and energy consumption data can be obtained through multivariate state estimation technology. In the fault list of Table 1, each group of fault codes represents all fault types that occurred in the device within a period of time.

[0069] Fault list Fault list G1, G2 G2, G4, G7 G1, G2, G3, G8 G3, G4, G5 G3, G5, G6 G1, G3, G4 G3, G7 G4, G5 G1, G8 G3, G6

[0070] Table 1

[0071] Next, the data is filtered twice. The first filter will remove the item sets with support counts less than 2, and the second filter will remove the item sets with support counts less than 1. Table 2 shows the first filter.

[0072] Itemset Support Count G1 4 G2 3 G3 6 G4 4 G5 3 G6 2 G7 2 G8 2

[0073] Table 2

[0074] Table 3 is the second screening;

[0075] Itemset Support Count G3 → G1 2 G3 → G2 1 G3 → G4 2 G3 → G5 2 G3 → G6 2 G3 → G7 1 G3 → G8 1

[0076] Table 3

[0077] From the screening results, we can find that the support counts of all the item sets in the first screening are greater than 2, so they are not eliminated. At the same time, fault code G3 is the most frequent fault type. In the second screening, the correlation between G3 and other faults is analyzed. The confidence and relevance of the item sets after the second screening are as follows: Figure 2 shown.

[0078] Among the seven item sets {G3→G1}, {G3→G2}, {G3→G4}, {G3→G5}, {G3→G6}, {TG3→G7}, and {G3→G8}, the correlation L of three item sets is greater than 1, which means that the occurrence of fault G3 will increase the possibility of the occurrence of faults G1, G5, and G6; the correlation L of four item sets is less than 1, that is, the occurrence of fault G3 will inhibit the occurrence of G2, G4, G7, and G8. The size of the possibility is given by the confidence C. Assuming that the time scale of each group of fault codes in the fault database is one week, when fault G3 occurs, the probability of faults G1, G5, and G6 occurring in the next week is 33.33%.

[0079] In summary, the Apriori algorithm can be used to implement an association rule system to predict possible failures and issue warnings. In this way, enterprises can detect equipment failures in advance and take corresponding measures to avoid production losses caused by equipment failures, providing an effective scientific basis for equipment troubleshooting and maintenance, and improving the reliability and stability of equipment.

[0080] The contents described in the embodiments of this specification are merely enumerations of implementation forms of the inventive concept and are for illustrative purposes only. The protection scope of the present invention should not be considered to be limited to the specific forms described in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be thought of by ordinary technicians in this field based on the inventive concept.

Claims

1. A method for analyzing equipment failure based on energy consumption big data, characterized in that: The method comprises the following steps: 1) Data collection: Collect and aggregate energy consumption data of devices, including power consumption levels, voltage fluctuations, and current usage information, using sensors and smart meters to capture real-time energy consumption data; 2) Data preprocessing: Clean and preprocess the raw data to handle missing values, outliers, and inconsistencies; convert the data into a set format for analysis to ensure compatibility with the selected analysis method; extract relevant features from the energy consumption data to gain in-depth understanding of device performance, including average power consumption, peak demand, and power peak frequency; 3) Big Data Analysis: Apply association rules and multivariate state estimation techniques to learn equipment operation energy consumption data to identify patterns, trends, and anomalies in energy consumption data; classify normal and abnormal equipment behaviors, and use historical data to train models to identify patterns associated with various types of failures; 4) Setting thresholds: Define safety thresholds for different parameters based on the energy consumption data when the equipment is working normally. When the actual data deviates significantly from the safety threshold, it may indicate a potential equipment failure. 5) Alarm generation: When the analysis identifies a potential equipment failure, a notification or alarm is generated through the alarm system, which is sent to maintenance personnel or system operators for further investigation.

2. The equipment failure analysis method based on energy consumption big data according to claim 1, characterized in that: In step 3), the processing process of the multivariate state estimation technology MSET is as follows: MSET is a nonlinear multivariate predictive diagnosis technology, which calculates and estimates various parameters during normal operation by analyzing and comparing actual monitoring parameters with health data during normal operation of the equipment, and uses this normal state estimation as a standard; When the actual operation data is obtained, the health data is also used as a basis, and the degree of correlation between the actual data and the health data is found to estimate the actual operation status; The degree of association is determined by the weight vector, which is used to measure the similarity between the actual state and the normal state. Finally, the estimated results of the healthy state and the actual operating state are compared and analyzed, and the concept of residual is introduced, and finally the equipment is diagnosed.

3. A method for analyzing equipment failure based on energy consumption big data as described in claim 1 or 2, characterized in that: In the step 3), the association rules are as follows: the association rules are in the form of "if-then" statements, which represent the relationship between different items in the data set. Based on the multivariate state estimation technology, the Apriori association rules are used to mine the association between the energy consumption data when the equipment is in different fault states, and the association between the most common faults in the equipment and other faults is studied. Based on this, early warnings are given for possible faults, and corresponding maintenance suggestions are given to maintenance personnel.

4. The equipment failure analysis method based on energy consumption big data as claimed in claim 3 is characterized in that: The association rule is based on the existence of a certain regularity between the values ​​of two or more variables. This regularity is called association. The association rule is expressed as an implication formula in the form of X→Y, where X and Y are respectively called the predecessor and successor of the association rule. The support, confidence and relevance of the association rule X→Y are used to measure the strength and reliability of the rule. Assume that there is a transaction data set P = {p1, p2, p3, ..., p k }, this transaction data set has k data, and the events represented by X and Y belong to this set; The support S of the association rule X→Y is defined as the proportion of X∪Y tuples contained in the data set P, which can be expressed in the following form using probability theory: Represents the probability of item sets X and Y occurring simultaneously; The confidence C of the association rule X→Y represents the probability that item set Y will occur if item set X occurs: The relevance L of the association rule X→Y is the ratio of the confidence C to the support S, which reflects how much the probability of the occurrence of X changes the occurrence of Y: Among them, the correlation Lift(X→Y) is the most important value in the association rule. It indicates the strength of the association between events X and Y, and represents the correlation between fault X and other faults Y. When Lift(X→Y)>1, it means that the occurrence of fault X will increase the occurrence of fault Y; when Lift(X→Y)<1, it means that the occurrence of fault X will inhibit the occurrence of fault Y.

5. The equipment failure analysis method based on energy consumption big data as claimed in claim 4 is characterized in that: The association rule mining process comprises the following steps: 3.1) Data preparation: This involves collecting and preprocessing the dataset to ensure that it is in a format suitable for analysis; 3.2) Frequent Itemset Generation: A frequent itemset is a group of items (e.g., energy consumption data when a device has different types of failures) that appear together in a sufficient number of instances in the dataset. The Apriori algorithm is used to identify frequent itemsets. 3.3) Association rule generation: Once the frequent itemsets are identified, association rules are generated from them. The association rules take the form of "if {X} then {Y}", where X and Y are sets of items representing different types of equipment failures; 3.4) Rule evaluation and pruning: In this step, the generated association rules are evaluated based on their interestingness measures (e.g., support, confidence, relevance) and may be pruned to remove rules that do not meet certain criteria or are not considered interesting; 3.5) Interpretation and Visualization: The discovered association rules are interpreted and visualized to gain insight into the relationships between items in the dataset.

Images