A method and system for visual analysis of abnormal patterns in machine tool operation data

Abstract

The present invention discloses a visual analysis method and system for the association of abnormal patterns of machine tool operation data. The method extracts symbol features from the sliding window and time points of the original machine tool operation data, performs multi-parameter category coding, obtains a multi-parameter coding matrix sequence, extracts the multi-parameter category coding combination of each column as a data sample, performs cluster analysis and association rule mining, extracts the machine tool operation mode according to the cluster analysis, and extracts the parameter state association pattern according to the association rule set obtained by association rule mining; constructs an interactive visual analysis system to realize visual analysis mapping of multi-parameter category coding, machine tool operation mode and parameter state association pattern, and designs multiple interactive methods to support users in discovering abnormal patterns of machine tool data, analyzes abnormal changes of various variables in machine tool data and captures the association relationship between various variables, can quickly discover and analyze anomalies, and fully mines the mutual influence, association and hidden correlation information between machine tool parameters.

Description

Technical Field

[0001] The present invention relates to the field of abnormal event detection of CNC machine tool operation history data, and in particular to a method and system for visual analysis of abnormal pattern association of machine tool operation data. Background Art

[0002] CNC machine tools play a vital role in the industrial sector. They can control machine movement through digital instructions and complete a variety of complex parts processing tasks. They are widely used in military equipment, automobile manufacturing, mold processing, aerospace, and other fields. However, frequent failures of machine tools during operation can lead to multiple problems, including damage to the equipment itself, reduced production efficiency, economic losses, and the occurrence of safety accidents.

[0003] With advances in technology, time series symbolization technology provides effective methods for analytical tasks such as classification, clustering, and association rule mining of time series data. Furthermore, visual representation and analysis have gained widespread application across many industries, enabling them to better organize, manage, and understand their data. A variety of sensors and recorders can record and monitor the production and operation status of machine tools in real time. By analyzing and processing this data, users can quickly and accurately identify fault points and provide appropriate solutions. This not only reduces maintenance downtime, improves production efficiency, but also reduces maintenance costs. Therefore, effective monitoring, analysis, and maintenance of machine tool operation data are crucial.

[0004] Although some machine learning-based models can currently analyze and mine multidimensional time series data and perform well in detecting anomalies or ranges, they usually only identify anomalies or ranges in the data and provide some statistical information to describe these anomalies. They lack analysis of abnormal changes in each variable and capture the correlation between variables. The information provided in explaining the causes of anomalies is very limited, making it impossible to quickly and accurately analyze the causes of anomalies. Summary of the Invention

[0005] In response to the problems existing in the above-mentioned fields, the present invention provides a visual analysis method and system for the association of abnormal patterns in machine tool operation data, which can solve the technical problems of lack of analysis of abnormal changes in various variables and capture of the correlation relationship between various variables, very limited information provided in explaining the causes of abnormalities, and inability to quickly and accurately analyze the causes of abnormalities.

[0006] To solve the above technical problems, the present invention discloses a method and system for visual analysis of abnormal patterns in machine tool operation data, comprising the following steps:

[0007] Symbolic features are extracted from the sliding window and time points of the original machine tool operation data, and multi-parameter category encoding is performed to obtain a multi-parameter encoding matrix sequence;

[0008] According to the multi-parameter coding matrix sequence, the multi-parameter category coding combination of each column is extracted as a data sample, and cluster analysis and association rule mining are performed. The machine tool operation mode is extracted based on the clustering results obtained by cluster analysis, and the parameter state association pattern is extracted based on the association rule set obtained by association rule mining.

[0009] Build an interactive visual analysis system that supports users in discovering abnormal patterns in machine tool data by designing multiple interactive methods. This system can achieve visual analysis and mapping of multi-parameter category coding, machine tool operation mode, and parameter status association patterns, analyze abnormal changes in various variables in machine tool data, and capture the associations between these variables.

[0010] The step of extracting symbol features from the sliding window and time points of the original machine tool operation data comprises the following steps:

[0011] The z-score standardization is used for multiple subsequences of the original machine tool operation data, and each subsequence is standardized to a subsequence with a mean of 0 and a standard deviation of 1; the subsequence S = {t1, t2, t3, t4, ..., t n}, the normalized sequence s is:

[0012]

[0013] Where μ is the mean value of all elements in S, and σ is the standard deviation of the sequence;

[0014] The standardized multidimensional time series matrix C is:

[0015]

[0016] Where n is the length of the time series, and m is the number of machine tool parameters;

[0017] The multidimensional time series matrix C is segmented according to the fixed sliding window segmentation method, w is the length of the sliding window, and the resulting encoding matrix sequence D is:

[0018]

[0019] Among them, n-w+1 is the number of sliding windows, d ij is the sliding window average; s ij The data after normalization of the i-th parameter at the j-th time point;

[0020] According to the breakpoint table, the sliding window average d ij The symbols are encoded and the encoding matrix E is obtained as:

[0021]

[0022] Among them, e ij Encode the sign of the mean.

[0023] Preferably, the z-score standardization is used on the multiple subsequences of the original machine tool operation data to standardize each subsequence, comprising the following steps:

[0024] Encode multiple parameter categories at each time point, and obtain the encoding matrix F as follows:

[0025]

[0026] Among them, m represents the machine tool parameters, n represents the time series length, f ij is the symbol code of the time point; for the code of the multi-parameter category of the time point, it is represented by letters representing different regional divisions and numbers representing different parameters, that is, the code combination of the first column {f 11 ,f 21 ,f 31 ,f 41}={d1, c2, b3, a4}, indicating that at the first second, the first parameter is in state d; the second and third parameters are in states c and b respectively; and the fourth parameter is in state a.

[0027] Preferably, the extraction algorithm for extracting the machine tool operation mode includes the following steps:

[0028] Input: original machine tool operation data;

[0029] Output: type of machine tool operation mode;

[0030] Step 1: Perform z-score standardization on the original machine tool operation data;

[0031] Step 2: Segment the original data using the fixed sliding window segmentation method;

[0032] Step 3: Calculate the average value of the sliding window data, encode it with the letters corresponding to the divided areas, and construct the encoding matrix sequence E, where each column represents multiple parameter states of a sliding window;

[0033] Step 4: Extract the multi-parameter category coding combinations of each column of the coding matrix E, remove the duplicates of the multi-parameter category coding combinations and perform clustering;

[0034] Step 5: According to the clustering results, each sliding window encoding combination is divided into different clusters to express various machine tool operation modes.

[0035] Preferably, the association rule extraction algorithm for association rule mining comprises the following steps:

[0036] Input: original machine tool operation data;

[0037] Output: machine tool parameter status association rule set;

[0038] Step 1: Perform z-score standardization on the original machine tool operation data;

[0039] Step 2: Encode the data at each time point. The encoding uses letters representing different area divisions and numbers representing different parameters.

[0040] Step 3: Construct the encoding matrix F, and combine the encoding results of each time point into a matrix by column to obtain a matrix containing multiple columns, where each column represents multiple parameter states at a time point;

[0041] Step 4: Extract the multi-parameter category coding combination of each column of the coding matrix F, apply the Apriori association rule algorithm to mine, and set the support and confidence thresholds;

[0042] Step 5: According to the set support and confidence thresholds, the association rule set between the machine tool parameter states is obtained.

[0043] Preferably, the setting of support and confidence thresholds comprises the following steps:

[0044] The support calculation formula is:

[0045]

[0046] Among them, support(·) represents support; represents the relative probability of a and d occurring simultaneously in all states;

[0047] The confidence calculation formula is:

[0048]

[0049] Among them, confidence(·) represents the confidence level; It represents the probability of state d appearing after state a appears in all states.

[0050] Preferably, the algorithm for extracting the parameter state association pattern comprises the following steps:

[0051] Extract symbolic features from the time points of the original machine tool operation data and perform parameter category coding;

[0052] Mining association rules for multiple parameter coding combinations;

[0053] Multiple parameter state association patterns are expressed according to a set of association rules.

[0054] Preferably, a visual analysis system for correlation of abnormal patterns of machine tool operation data is also included, comprising:

[0055] A data acquisition module is used to extract symbolic features from the sliding window and time points of the original machine tool operation data;

[0056] The data processing module performs multi-parameter category encoding based on the symbolic features extracted by the data acquisition module to obtain a multi-parameter encoding matrix sequence; and designs multiple interactive methods to support users in discovering abnormal patterns in machine tool data;

[0057] The data analysis module extracts the multi-parameter category coding combination of each column as a data sample based on the multi-parameter coding matrix sequence obtained by the data processing module, performs cluster analysis and association rule mining, extracts the machine tool operation mode based on the clustering results obtained by the cluster analysis, and extracts the parameter state association pattern based on the association rule set obtained by the association rule mining;

[0058] The data visual analysis module is used to build an interactive visual analysis system, analyze the abnormal patterns of the data processing module, realize the visual analysis mapping of multi-parameter category coding, machine tool operation mode and parameter status association pattern, obtain abnormal changes of each variable in the machine tool data and capture the association relationship between each variable.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. The interactive visual analysis system constructed by the present invention embeds machine tool operation mode and parameter state association pattern extraction algorithms, supports users to flexibly explore data attribute changes, and provides methods such as anomaly location, sensitive parameter analysis, state similarity search, and parameter state association analysis for operation data monitoring, intuitively and efficiently extracting potential information from production process data. Users can use the method designed by the present invention to conduct comprehensive interactive analysis of fault data.

[0061] 2. By encoding the parameter categories of machine tool operation data from two aspects, namely sliding windows and time points, and extracting symbolic features, the system can support users in discovering abnormal patterns in machine tool data based on a variety of designed interactive methods. By analyzing the abnormal changes of various variables in machine tool data and capturing the correlation between the variables, it is possible to quickly discover and analyze which parameters cause abnormalities.

[0062] 3. In combination with the mechineVis visual analysis system, a case study was conducted on a machine tool dataset through visual analysis mapping and interactive linkage of multiple views. The results demonstrated that the visual analysis system of the present invention can easily discover and analyze anomalies and fully explore the mutual influence, association, and hidden correlation information among machine tool parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of the visual analysis system mechineVis of the present invention;

[0064] Figure 2 This is a schematic diagram of sliding window multi-parameter category encoding of the present invention;

[0065] Figure 3 This is a schematic diagram of the multi-parameter coding combination clustering of the present invention;

[0066] Figure 4 This is a schematic diagram of multi-parameter category coding at a time point of the present invention;

[0067] Figure 5 This is a schematic diagram of mining association rules of time point multi-parameter coding combinations according to the present invention;

[0068] Figure 6 This is a diagram of the mechineVis visual analysis system of the present invention;

[0069] Figure 7 The analysis result diagram is a global view of the embodiment part of the present invention;

[0070] Figure 8 This is a diagram showing the analysis results of the key indicator view of the embodiment of the present invention;

[0071] Figure 9 A diagram showing analysis results of a partial view of an embodiment of the present invention;

[0072] Figure 10 This is a diagram showing the analysis results of the parameter state association view of the embodiment of the present invention;

[0073] Figure 11 This is a diagram showing the analysis results of abnormal areas in the embodiment of the present invention;

[0074] Figure 12 This is a diagram showing the analysis results of abnormal patterns in the embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following is a combination of the embodiments of the present invention Figures 1-12 , the technical solutions in the embodiments of the present invention are clearly and completely described. It should be understood that the terms used in the present invention are only used to describe specific implementation methods and are not intended to limit the present invention.

[0076] In recent years, various feature extraction techniques have been developed in time series mining, including symbolization techniques. Symbolic clustering is a popular method. This method first segments the time series based on the segmented clustering method and uses the mean of each segment to represent the original data. The original data is then quantized into multiple regions according to probability using a standard normal distribution. Symbols represent the values of different regions, and the means are mapped to the symbols. Regarding visual analysis for anomaly detection, automated anomaly detection methods often overlook subtle, ambiguous, and uncertain anomalies. However, visual analysis improves data analysis capabilities through visual representation and interfaces. Time series symbolization techniques extract feature information, providing effective methods and tools for tasks such as classification, clustering, association rule mining, and similarity measurement of time series data. Visual representation and analysis have been widely used in many industries, providing them with the ability to better organize, manage, and understand industry data. However, research on visual analysis for anomaly detection in machine tool fault diagnosis is relatively limited.

[0077] The visual analysis system mechineVis proposed in the present invention realizes visual analysis mapping of multi-parameter category coding, machine tool operation mode, and parameter status association mode, and designs multiple interactive methods to support users in discovering abnormal patterns in machine tool data, analyzing the state changes of various parameters in machine tool data and capturing the association between various parameters, helping users understand machine tool operation data and analyze the causes of abnormalities, quickly locate problems, reduce downtime, and improve production efficiency.

[0078] The present invention proposes a method for visual analysis of abnormal patterns in machine tool operation data, which is implemented by the following steps:

[0079] The first step is to encode and map the sliding window multi-parameter categories;

[0080] Extract symbolic features from the sliding window and time points of machine tool operation data and perform parameter category coding;

[0081] The z-score standardization is used on multiple subsequences of the original operating data of the motor bearing, and each subsequence is standardized to a subsequence with a mean of 0 and a standard deviation of 1. Therefore, for the subsequence S = {t1, t2, t3, t4, ..., t n}, the normalized sequence s is:

[0082]

[0083] Among them, μ is the average value of all elements in S, σ is the standard deviation of the sequence, that is, the standardized multidimensional time series matrix C is obtained, which is expressed as:

[0084]

[0085] Among them, n is the length of the time series and m is the number of machine tool parameters.

[0086] The multidimensional time series matrix C is segmented according to the fixed sliding window segmentation method, w is the length of the sliding window, and the encoding matrix sequence D can be obtained, which is expressed as:

[0087]

[0088] Among them, n-w+1 is the number of sliding windows, d ij is the sliding window average; s ij The data after normalization of the i-th parameter at the j-th time point;

[0089] Sliding window average d ij The symbol encoding is mapped according to the discontinuity table, as shown in Table 1, and the encoding matrix E is obtained as follows:

[0090]

[0091] Among them, e ij Encode the sign of the mean.

[0092] Table 1 Breakpoint table

[0093]

[0094] According to Table 1, when A=4, there are three breakpoints, namely -0.67, 0, and 0.67, which are divided into 4 regions. The part with a value less than -0.67 is represented by the letter a, the part with a value between -0.67 and 0 is represented by the letter b, the part with a value between 0 and 0.67 is represented by the letter c, and the part with a value greater than 0.67 is represented by the letter d. Figure 2 shown.

[0095] The second step is to cluster multiple parameter coding combinations;

[0096] For the encoding matrix sequence Extract the multi-parameter category coding combination of each column as data sample for cluster analysis, such as Figure 3 As shown in the figure, the steps of the machine tool operation mode extraction algorithm are as follows:

[0097] Input: Original machine tool operation data

[0098] Output: Multiple operation modes of machine tools

[0099] Step 1: Perform z-score standardization on the original machine tool operation data

[0100] Step 2: Segment the original data using the fixed sliding window segmentation method

[0101] Step 3: Calculate the average value of the sliding window data, encode it with the letters of the corresponding divided areas, and construct the encoding matrix sequence E, where each column represents multiple parameter states of a sliding window

[0102] Step 4: Extract the multi-parameter category coding combinations of each column of the coding matrix E, remove the duplicates of the multi-parameter category coding combinations and perform clustering

[0103] Step 5: Divide each sliding window into different clusters based on the clustering results to express various machine tool operation modes.

[0104] Clustering divides the sample set into multiple clusters, ensuring that data within the same cluster is highly similar and data between different clusters are highly different. Clustering typically measures similarity by calculating the Euclidean distance from each sample point to the cluster center. However, the distances of the encoded letters in the sliding window data cannot be calculated numerically. This invention provides a distance table between letters, as shown in Table 2, and performs clustering by examining the distance table between each letter.

[0105] Table 2 Distance table

[0106]

[0107] The third step is to express various machine tool operation modes based on the clustering results;

[0108] Extract symbolic features from the time points of machine tool operation data and encode parameter categories; map the multi-parameter category codes at the time points; when mining machine tool operation data, in order to fully explore the relationship between various parameters, after standardizing the machine tool operation data, encode multiple parameter categories at each time point, and obtain the encoding matrix F:

[0109]

[0110] Among them, m represents the machine tool parameters, n represents the time series length, f ij A symbolic code for a time point.

[0111] For the coding of multiple parameter categories at a time point, letters representing different area divisions and numbers representing different parameters are used to represent them. For example, the coding combination in the first column {f 11 ,f 21 ,f 31 ,f 41}={d1,c2,b3,a4}, which means that at the first second, the first parameter is in state d; the second and third parameters are in states c and b respectively; the fourth parameter is in state a; the obtained time point multi-parameter category code is as follows Figure 4shown.

[0112] Mining association rules for multiple parameter coding combinations;

[0113] For each column of the coding matrix F, the Apriori association rule algorithm is applied to generate a set of state association rules between parameters. It essentially consists of two parts: support and confidence.

[0114] Support is the relative probability that two different states a and d with different parameters occur simultaneously, that is:

[0115]

[0116] Among them, support(·) represents support;

[0117] Confidence represents the probability of state d appearing when state a appears in all states, that is:

[0118]

[0119] Among them, confidence(·) represents the confidence level;

[0120] Users can independently set support and confidence thresholds based on actual needs and data characteristics to screen out association rules with certain significance and credibility.

[0121] like Figure 5 As shown in the figure, for n parameters in the machine tool operation data, m letters are used to encode their states. Therefore, each parameter has m states, for a total of m×n possible states. The support and confidence between different parameter states are calculated, and association rules that meet the threshold requirements are screened to discover the association relationships between the various parameter states.

[0122] The machine tool parameter state association rule extraction algorithm includes the following steps:

[0123] Input: Original machine tool operation data

[0124] Output: Machine tool parameter status association rule set

[0125] Step 1: Perform z-score standardization on the original machine tool operation data

[0126] Step 2: Encode the data at each time point. The encoding is represented by letters representing different area divisions and numbers representing different parameters.

[0127] Step 3: Construct the encoding matrix F, and combine the encoding results of each time point into a matrix by column. This will result in a matrix containing multiple columns, each column representing multiple parameter states at a time point.

[0128] Step 4: Extract the multi-parameter category coding combination of each column of the coding matrix F, apply the Apriori association rule algorithm to mine, and set the support and confidence thresholds

[0129] Step 5: According to the set support and confidence thresholds, the association rule set between the machine tool parameter states is obtained.

[0130] The present invention expresses multiple parameter state association patterns based on association rule sets, and aims to provide researchers with an interactive exploration tool through the detailed design and interaction of each view in the visual analysis system mechineVis system, by mapping the visual analysis of multi-parameter category encoding and embedding multiple pattern extraction algorithms. The mechineVis system mainly includes five key views, such as Figure 6 As shown, it specifically includes: (a) control panel view, (b) local view, (c) key indicator comparison view, (d) parameter status association view and (e) global view.

[0131] like Figure 6 As shown in Figure (a), the control panel view includes five options from top to bottom: data selection, number of clusters, sliding window length, support, and confidence. Data selection determines the machine tool operation data to be analyzed. The number of clusters and sliding window length are related to the extraction of machine tool operation modes. The number of clusters represents the number of machine tool operation modes, and the sliding window length represents the time series span of the machine tool operation data segmented by the sliding window. The support and confidence settings are related to the extraction of correlation patterns between machine tool parameter states. The larger the support and confidence, the stronger the correlation between parameter states.

[0132] like Figure 6 As shown in Figure (e), the global view displays machine tool operating data using a line graph. The shaded area of the line graph represents the magnitude and duration of data fluctuations. The background color at the mouse position represents the current machine tool operating status. When the user positions the mouse at a certain moment, the bearing operating status is represented by a different color. If the symbol contains four letters, the bearing operating status is represented by four colors. When the user clicks on a specific moment, the global view searches for the same time segment as the current machine tool operating status. Similarity searches are implemented by determining whether the code combinations are equal, thus matching the same operating status.

[0133] According to the clustering results of the sliding window multi-parameter coding combination, different clusters can represent different operating modes of the machine tool, and different clusters are mapped with different colors. Figure 6 In Figure (e), the rectangle at the bottom of the view shows changes in the machine's operating mode, providing an intuitive overview of the bearing's operating status. A change in color indicates a change in the machine's operating mode. Additionally, a time brush is provided in the global view, allowing you to adjust the display range of the local view using a slider or mouse drag.

[0134] like Figure 6 As shown in Figure (b), by interpreting the local view, users can more closely observe the specific area where the anomaly is located and conduct in-depth analysis of the cause of the anomaly. The left half of the local view displays the detailed changes in each parameter data over a specific time period. Each rectangle represents the current state. The color change position of the rectangle can be used to dig out detailed information about the operating data changes of each parameter.

[0135] The colored rectangular bar at the bottom of the local view represents the operating mode of the bearing during the selected time period. By distinguishing different operating modes by color, users can more easily analyze the operating data of various parameters under abnormal modes.

[0136] Also in Figure 6 Figure (b) in the figure provides a comparative analysis of the state at a certain moment and the previous state. Place the mouse on a certain moment, such as Figure 6 The two small rectangles on the left side of the (b) diagram will show the status of the previous second and the running status of the previous two seconds. The large rectangle represents the current running status of the mouse. If the colors of the large and small rectangles are different, it means that the status of the current parameter has changed compared to the previous two seconds. Figure 6 The heat map in the right half of Figure (b) represents the correlation between different parameters within the selected time period. Different colors in the heat map represent different correlations, such as red for a high positive correlation and purple for a high negative correlation. In addition, a correlation index filter is available in the local view.

[0137] like Figure 6 As shown in Figure (d), it is a parameter state association view. In the parameter state association view, the relationship between each parameter state in the machine tool operation data can be fully mined. After the association rules of the multi-parameter coding combination are mined, the filtered association rules are visually analyzed and displayed.

[0138] Each circle represents a state, with different colored circles representing different states, allowing users to quickly distinguish between different parameter states. The lines connecting the circles represent an association rule. The color of the line represents support, while the width of the line represents confidence. Darker lines indicate higher support, while wider lines indicate higher confidence. When users click on a state, they can analyze it in conjunction with the global view to discover the associations between different states of each parameter and the timing of the state associations.

[0139] like Figure 6 Figure (c) shows the key indicator comparison view. This view uses a radar chart to compare the characteristic values of various machine tool parameters under different operating modes. The compared characteristic values include mean, variance, and standard deviation. By comparing the characteristic values of multiple parameters under different operating modes, we can understand how the parameters change under different operating modes, thereby better evaluating the machine tool's operating status.

[0140] The present invention also proposes a visual analysis system for abnormal pattern correlation of machine tool operation data, comprising:

[0141] A data acquisition module is used to extract symbolic features from the sliding window and time points of the original machine tool operation data;

[0142] The data processing module performs multi-parameter category encoding based on the symbolic features extracted by the data acquisition module to obtain a multi-parameter encoding matrix sequence; and designs multiple interactive methods to support users in discovering abnormal patterns in machine tool data;

[0143] The data analysis module extracts the multi-parameter category coding combination of each column as a data sample based on the multi-parameter coding matrix sequence obtained by the data processing module, performs cluster analysis and association rule mining, extracts the machine tool operation mode based on the clustering results obtained by the cluster analysis, and extracts the parameter state association pattern based on the association rule set obtained by the association rule mining;

[0144] The data visual analysis module is used to build an interactive visual analysis system, analyze the abnormal patterns of the data processing module, realize the visual analysis mapping of multi-parameter category coding, machine tool operation mode and parameter status association pattern, obtain abnormal changes of each variable in the machine tool data and capture the association relationship between each variable.

[0145] The visual analysis system includes the mechineVis system, which is used to map the visual analysis of the multi-parameter category encoding obtained by the data analysis module and embed multiple pattern extraction algorithms of the data processing module.

[0146] Among them, the mechineVis system includes: control panel view, global view, local view, key indicator comparison view, and parameter status association view.

[0147] The present invention extracts symbolic features from the sliding window and time points of machine tool operation data and combines them into parameters, thereby encoding the machine tool operation status. The sliding window multi-parameter coding combinations are then clustered to identify different machine tool operation modes. Association rule mining is then performed on the time point multi-parameter coding combinations to calculate the support and confidence between each parameter state, and further analyze the association relationship in the data. Finally, combined with the visual analysis system mechineVis, a case study was conducted on a machine tool dataset through visual analysis mapping and multi-view interaction. This demonstrated that the visual analysis system of the present invention can easily discover and analyze anomalies and fully explore the mutual influence, association, and hidden correlation information between machine tool parameters.

[0148] Example

[0149] The present invention uses the public dataset SKAB, which is a multivariate time series collected from sensors installed on a motor test bench. It contains eight parameters, namely Accelerometer 1RMS (vibration acceleration), Accelerometer 2RMS (vibration acceleration), Current (current), Pressure (loop pressure), Temperature (engine temperature), Thermocouple (loop fluid temperature), Voltage (motor voltage), VolumeFlowRateRMS (circulation flow rate of the loop fluid). The data is recorded every second from 2020-02-0817:27:19 to 2020-02-0817:47:18, with a length of 1144. First, the data needs to be z-score standardized. In the control panel view, the number of letters is set to 4, the length of the sliding window is set to 30, and the number of clusters is 3. At the same time, the support is set to 0.25 and the confidence is set to 0.85.

[0150] like Figure 7As shown in the rectangle at the bottom of the global view, the machine tool's operating mode is divided into three modes: Mode 1, Mode 2, and Mode 3. These modes are mapped using different colors. It is found that at time 585, the machine tool's operating mode changed. In the Mode 2 area, the parameters A1RMS (Accelerometer 1RM), A2RMS (Accelero meter 2RM), and VRMS (VolumeFlowRate RMS) have large shaded areas, indicating that these parameters experienced significant and prolonged fluctuations during this time period, likely indicating an anomaly. Therefore, the distribution of these parameter states during this time period warrants special attention. Drag the time brush in the heat map at the bottom right of the global view to position the time range within the Mode 2 area. Observing the local view, we can see that the red areas of A1RMS and A2RMS cover a wide area, and the light blue area of VRMS also occupies a large area. This indicates that during this period, the A1RMS and A2RMS data were relatively high, while the VRMS data was mostly low. This further confirms that anomalies occurred in A1RMS, A2RMS, and VRMS during the time period of Mode 2. Observing the correlation heat map on the left of the local view, we can see that during this period, A1RMS and A2RMS were positively correlated, and there was also a high negative correlation between A1RMS, A2RMS, and VRMS.

[0151] like Figure 8 As shown in the key indicator view, the radar chart comparison shows that: in the three operating modes, the characteristic values of Current, Voltage, and Pressure are relatively close, indicating that Current, Voltage, and Pressure are in a stable state throughout the entire working process. However, there are large differences in the characteristic values of A1RMS, A2RMS, VRMS, Temperature, and Thermocouple in different operating modes. Figure 7 The main view indicates that A1RMS, A2RMS, and VRMS may be due to unstable machine operation or abnormal factors. Furthermore, Temperature and Thermocouple show an overall downward trend, leading to significant differences in eigenvalues between different modes.

[0152] like Figure 9As shown in the figure, a detailed analysis of the three abnormal sequences, A1RMS, A2RMS, and VRMS, in the local view reveals that the color arrangement of the A1RMS and A2RMS rectangles significantly changed before the machine abnormality occurred. Combined with the changes in the large and small rectangles and mouse positioning, it was found that A2RMS and A1RMS underwent significant changes in state starting at the 497th and 500th seconds, respectively. Similarly, after the machine abnormality occurred, VRMS began to gradually decrease from the 596th second, with a significant change in state compared to the previous period. Therefore, it can be inferred that the continued increase in A1RMS and A2RMS may have caused the machine abnormality, leading to the VRMS abnormality.

[0153] like Figure 10 As shown in the parameter state association view, it is found that A1RMS(d), A2RMS(d), and VRMS(a) are related to each other, and after clicking on one of the states, the position of each state can be displayed in the global view. Figure 11 As shown, it is found that the region where the correlation state of A1RMS(d), A2RMS(d), and VRMS(a) appears is within the abnormal range, that is, it can be inferred that A1RMS, A2RMS, and VRMS usually produce abnormalities together, and when A1RMS and A2RMS are high, VRMS is low.

[0154] A1RMS(a) and A2RMS(a) are also correlated, indicating that in Modes 1 and 3, when A1RMS is low, A2RMS is also low. Furthermore, Temperature(d) and Thermocouple(d) are found to be trending towards A1RMS(a), indicating that in Mode 1, when Temperature and Thermocouple are high, A1RMS is generally low.

[0155] like Figure 12 As shown in the figure, when you click the background at 769 seconds in the light red area of the global view, you can find that the same operating state also appeared at 775, 788, 865, 887, 896, 903, 930, and 952 seconds. The same abnormal pattern can be observed at the time of occurrence, indicating that there are many similar operating states in the interval [769-952].

[0156] The present invention achieves encoding of machine tool parameter categories by extracting symbolic features from the sliding window and time points of the original machine tool operation data and performing parameter combination. The sliding window multi-parameter coding combinations are clustered to identify different machine tool operation modes. The time point multi-parameter coding combinations are then subjected to association rule mining to calculate the support and confidence between each parameter state, and further analyze the association relationship in the data. Combined with the interactive visual analysis system mechineVis, a case study was conducted on a machine tool dataset through visual analysis mapping and interactive linkage of multiple views, demonstrating that the visual analysis system of the present invention can easily discover and analyze which parameters cause anomalies, and can fully explore the mutual influence, association and hidden correlation information between machine tool parameters.

[0157] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

[0158] In addition, unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those skilled in the art to which the present invention belongs. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods related to the documents. In the event of any conflict with any incorporated document, the content of this specification shall prevail.

Claims

1. A visual analysis method for abnormal pattern correlation of machine tool operation data, characterized in that: The following steps are involved: Symbolic features are extracted from the sliding window and time points of the original machine tool operation data, and multi-parameter category encoding is performed to obtain a multi-parameter encoding matrix sequence; According to the multi-parameter coding matrix sequence, the multi-parameter category coding combination of each column is extracted as a data sample, and cluster analysis and association rule mining are performed. The machine tool operation mode is extracted based on the clustering results obtained by cluster analysis, and the parameter state association pattern is extracted based on the association rule set obtained by association rule mining. Build an interactive visual analysis system that supports users in discovering abnormal patterns in machine tool data by designing multiple interactive methods. This system can achieve visual analysis and mapping of multi-parameter category coding, machine tool operation mode, and parameter status association patterns, analyze abnormal changes in various variables in machine tool data, and capture the associations between these variables. The step of extracting symbol features from the sliding window and time points of the original machine tool operation data comprises the following steps: The z-score standardization is used for multiple subsequences of the original machine tool operation data, and each subsequence is standardized to a subsequence with a mean of 0 and a standard deviation of 1; the subsequence S = {t1, t2, t3, t4, ..., t n }, the normalized sequence s is: Where μ is the mean value of all elements in S, and σ is the standard deviation of the sequence; The standardized multidimensional time series matrix C is: Where n is the length of the time series, and m is the number of machine tool parameters; The multidimensional time series matrix C is segmented according to the fixed sliding window segmentation method, w is the length of the sliding window, and the resulting encoding matrix sequence D is: Among them, n-w+1 is the number of sliding windows, d ij is the sliding window average; s ij The data after normalization of the i-th parameter at the j-th time point; According to the breakpoint table, the sliding window average d ij The symbols are encoded and the encoding matrix E is obtained as: Among them, e ij Encode the sign of the mean.

2. A visual analysis method for abnormal pattern correlation of machine tool operation data according to claim 1, characterized in that: The method of using z-score to normalize multiple subsequences of the original machine tool operation data and performing normalization processing on each subsequence includes the following steps: Encode multiple parameter categories at each time point, and obtain the encoding matrix F as follows: Among them, m represents the machine tool parameters, n represents the time series length, f ij is the symbol code of the time point; for the code of the multi-parameter category of the time point, it is represented by letters representing different regional divisions and numbers representing different parameters, that is, the code combination of the first column {f 11 ,f 21 ,f 31 ,f 41 }={d1, c2, b3, a4}, indicating that at the first second, the first parameter is in state d; the second and third parameters are in states c and b respectively; and the fourth parameter is in state a.

3. The method for visually analyzing abnormal patterns of machine tool operation data according to claim 2, characterized in that: The extraction algorithm for extracting the machine tool operation mode includes the following steps: Input: original machine tool operation data; Output: type of machine tool operation mode; Step 1: Perform z-score standardization on the original machine tool operation data; Step 2: Segment the original data using the fixed sliding window segmentation method; Step 3: Calculate the average value of the sliding window data, encode it with the letters corresponding to the divided areas, and construct the encoding matrix sequence E, where each column represents multiple parameter states of a sliding window; Step 4: Extract the multi-parameter category coding combinations of each column of the coding matrix E, remove the duplicates of the multi-parameter category coding combinations and perform clustering; Step 5: According to the clustering results, each sliding window encoding combination is divided into different clusters to express various machine tool operation modes.

4. The method for visually analyzing abnormal patterns of machine tool operation data according to claim 3, characterized in that: The association rule extraction algorithm for association rule mining includes the following steps: Input: original machine tool operation data; Output: machine tool parameter status association rule set; Step 1: Perform z-score standardization on the original machine tool operation data; Step 2: Encode the data at each time point. The encoding uses letters representing different area divisions and numbers representing different parameters. Step 3: Construct the encoding matrix F, and combine the encoding results of each time point into a matrix by column to obtain a matrix containing multiple columns, where each column represents multiple parameter states at a time point; Step 4: Extract the multi-parameter category coding combination of each column of the coding matrix F, apply the Apriori association rule algorithm to mine, and set the support and confidence thresholds; Step 5: According to the set support and confidence thresholds, the association rule set between the machine tool parameter states is obtained.

5. The method for visually analyzing abnormal patterns of machine tool operation data according to claim 4, characterized in that: The setting of support and confidence thresholds includes the following steps: The support calculation formula is: Among them, support(·) represents support; represents the relative probability of a and d occurring simultaneously in all states; The confidence calculation formula is: Among them, confidence(·) represents the confidence level; It represents the probability of state d appearing after state a appears in all states.

6. The method for visual analysis of abnormal patterns in machine tool operation data according to claim 5, characterized in that: The extraction algorithm for extracting the parameter state association pattern comprises the following steps: Extract symbolic features from the time points of the original machine tool operation data and perform parameter category coding; Mining association rules for multiple parameter coding combinations; Multiple parameter state association patterns are expressed according to a set of association rules.

7. A visual analysis system for abnormal patterns of machine tool operation data, characterized in that: include: A data acquisition module is used to extract symbolic features from the sliding window and time points of the original machine tool operation data; The data processing module performs multi-parameter category encoding based on the symbolic features extracted by the data acquisition module to obtain a multi-parameter encoding matrix sequence; and designs multiple interactive methods to support users in discovering abnormal patterns in machine tool data; The data analysis module extracts the multi-parameter category coding combination of each column as a data sample based on the multi-parameter coding matrix sequence obtained by the data processing module, performs cluster analysis and association rule mining, extracts the machine tool operation mode based on the clustering results obtained by the cluster analysis, and extracts the parameter state association pattern based on the association rule set obtained by the association rule mining; The data visual analysis module is used to build an interactive visual analysis system to analyze abnormal patterns in the data processing module, realize visual analysis mapping of multi-parameter category coding, machine tool operation mode and parameter status association pattern, obtain abnormal changes in various variables of machine tool data and capture the association relationship between various variables; The step of extracting symbol features from the sliding window and time points of the original machine tool operation data comprises the following steps: The z-score standardization is used for multiple subsequences of the original machine tool operation data, and each subsequence is standardized to a subsequence with a mean of 0 and a standard deviation of 1; the subsequence S = {t1, t2, t3, t4, ..., t n }, the normalized sequence s is: Where μ is the mean value of all elements in S, and σ is the standard deviation of the sequence; The standardized multidimensional time series matrix C is: Where n is the length of the time series, and m is the number of machine tool parameters; The multidimensional time series matrix C is segmented according to the fixed sliding window segmentation method, w is the length of the sliding window, and the resulting encoding matrix sequence D is: Among them, n-w+1 is the number of sliding windows, d ij is the sliding window average; s ij The data after normalization of the i-th parameter at the j-th time point; According to the breakpoint table, the sliding window average d ij The symbols are encoded and the encoding matrix E is obtained as: Among them, e ij Encode the sign of the mean.

Images