Intelligent processing method and system for equipment health monitoring data

Abstract

The invention belongs to the technical field of intelligence, and discloses an equipment health monitoring data intelligent processing method and system. Comprising the steps of designing a dynamic acquisition mechanism and acquiring equipment state data; based on the ranking vector of the equipment state data at each moment and the data value vector of the equipment state data at each moment, obtaining comprehensive influence data at each moment by using a rank value mapping transformation measurement method; dividing the collected equipment state data into d groups according to the comprehensive influence data; based on the equipment state data in each group, calculating the comprehensive complexity of the group by using a dynamic mutation detection method; calculating a comprehensive complexity threshold value by using a random candidate-Youden maximization method to analyze the health state of the equipment at the moments included in the group; the intelligence and precision of equipment management are improved, the risk of equipment fault occurrence is reduced, and the overall operation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent technology, and in particular to a method and system for intelligent processing of equipment health monitoring data. Background Art

[0002] Existing equipment health monitoring data processing methods have many problems:

[0003] First, existing equipment health monitoring methods typically use a fixed data collection mode, always collecting the same type and frequency of data regardless of changing environmental conditions. This inability to respond to changes in environmental conditions in a timely manner results in insufficient equipment status information being obtained at critical moments, thus impacting equipment health monitoring capabilities. Alternatively, large amounts of data are collected without targeted filtering, resulting in inefficient data processing and analysis.

[0004] Secondly, existing methods predict the health status of equipment based on data at a single time point (such as vibration, temperature, current, etc.) or simple statistics (such as mean, standard deviation). They only focus on the data point at the current moment and ignore the evolution of the equipment status over time. Moreover, the data at a single time point is easily affected by measurement errors or accidental factors, which leads to inaccurate judgment of the equipment health status.

[0005] In view of this, the present invention proposes a method and system for intelligent processing of equipment health monitoring data to solve the above problems. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention provides the following technical solution: a method for intelligent processing of equipment health monitoring data, comprising:

[0007] Step SS1: Design a dynamic collection mechanism to collect device status data for k consecutive moments;

[0008] Step SS2: Use the time series ranking method to obtain the ranking vector of the device status data at each moment;

[0009] Step SS3: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, a rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment;

[0010] Step SS4: Using the comprehensive impact data at the initial moment as a benchmark, compare the comprehensive impact data at subsequent moments using the Euclidean distance, and set a distance threshold. When the comprehensive impact data at moment i exceeds the threshold compared to the current benchmark, the benchmark is updated, and the device status data corresponding to the previous comprehensive impact data are grouped together. Repeat this process until all device status data at all moments have been processed, and the device status data for k consecutive moments are divided into d groups.

[0011] Step SS5: Based on the device status data in each group, a dynamic mutation detection method is used to calculate the overall complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexity of all mutation points in the group is averaged to obtain the overall complexity of the group.

[0012] Step SS6: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

[0013] Furthermore, the specific methods of designing the dynamic collection mechanism include:

[0014] The dynamic collection mechanism includes basic mode and enhanced mode. The device status data collected in basic mode only includes key status data;

[0015] The device status data collected in enhanced mode includes key status data and auxiliary status data;

[0016] The trigger condition of the basic mode is always mandatory acquisition, which is not affected by other factors;

[0017] The trigger condition of the enhanced mode is dynamic activation on demand. When the ambient temperature is higher than the standard ambient temperature, the enhanced mode is triggered, and the collected device status data includes key status data and auxiliary status data.

[0018] Furthermore, the key status data includes bearing vibration data, gearbox vibration data, motor vibration data, motor temperature data, winding temperature data, lubricating oil temperature data, hydraulic system pressure data, pneumatic system pressure data, cooling system pressure data, motor current data, and motor voltage data.

[0019] Furthermore, when the ambient temperature is higher than the standard ambient temperature, the device status data includes key status data and auxiliary status data, wherein the auxiliary status data includes motor vibration acceleration data and vibration velocity data.

[0020] Furthermore, the specific method of using the time series ranking method to obtain the ranking vector of the device status data at each moment includes:

[0021] Retrieve the device status data at k moments, where the device status data includes key status data and auxiliary status data, or only key status data. If the device status data only includes key status data, the data values of the auxiliary status data at that moment are all assumed to be 0.

[0022] Sort the data values in the nth dimension at k moments in ascending order, with the smallest data value ranked 1 and the largest data value ranked k;

[0023] Calculate the ranking of the data values of each dimension at each moment and generate a ranking vector;

[0024] Data points with the same data value are given the same rank.

[0025] Furthermore, the specific method of using the rank value mapping transformation measurement method to obtain the comprehensive impact data at each moment includes:

[0026] The device status data at each moment includes n data points, that is, n dimensions. Each data point has a specific data value and ranking; therefore, each moment has a data value vector and ranking vector.

[0027] Using a normalization method to scale the data values in the data value vector to [0, 1], thereby obtaining scaled data values; performing a logarithmic transformation based on the scaled data values, thereby obtaining logarithmic transformed data values;

[0028] Use the normalization method to scale the ranking in the ranking vector to [0, 1] to obtain the scaled ranking; perform exponential transformation based on the scaled ranking to obtain the exponentially transformed ranking;

[0029] The influence data of a data point is obtained by performing a weighted summation of the logarithmic transformed data value and the exponential transformed ranking of the data point;

[0030] Repeat the calculation to obtain the impact data of n data points;

[0031] The average method is used to average the impact data of n data points to obtain the comprehensive impact data.

[0032] Furthermore, the specific method of using a dynamic mutation detection method to calculate the comprehensive complexity of the group includes:

[0033] Step a1: for the dth group, the group includes device status data at h moments;

[0034] Step a2: Use the Kalman filter method to smooth the data values of the device status data at h moments to obtain the filtered values of the device status data at h moments, and calculate the residual between the filtered values of the device status data at h moments and the data values of the device status data at h moments; if the residual exceeds the dynamic threshold, the data point corresponding to the data value is a mutation point;

[0035] Step a3: For the mutation point, use the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point;

[0036] Step a4: averaging the complexity of all mutation points in the dth group to obtain the comprehensive complexity index of the dth group.

[0037] Furthermore, the specific method of using the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point includes:

[0038] Step c1: When a mutation point is detected, a disturbance point is added to the data value of the mutation point to form a disturbance initial value; for the disturbance point, H disturbance points are randomly generated using a random method, and corresponding to H disturbance initial values;

[0039] Step c2: For each generated initial perturbation value, compare it with the data value of the mutation point to obtain the initial error;

[0040] Step c3: Set the number of iterations and the linear function, and use the linear function to iteratively calculate the mutation point and its initial perturbation value to obtain the trajectory of the mutation point and the initial perturbation value;

[0041] Step c4: In each iteration, the difference between the mutation point trajectory and the perturbation initial value trajectory is calculated;

[0042] Step c5: Take the natural logarithm of the relative error in each iteration to obtain a logarithm value. The relative error is the ratio of the difference between the mutation point trajectory and the perturbation initial value trajectory in the iteration to the initial error. After the iteration is completed, use the averaging method to calculate the average of the logarithm values of all iterations, which is the local Lyapunov exponent indicator.

[0043] Step c6: Using steps c1 to c5, calculate the local Lyapunov exponent index for H disturbance points; select the disturbance point corresponding to the maximum value of the local Lyapunov exponent index as the optimal disturbance point;

[0044] Step c7: Using the optimal perturbation point, calculate the mutation point trajectory and the optimal perturbation point trajectory, and calculate the logarithmic value of each iteration. After the iteration is completed, use the averaging method to calculate the average of the logarithmic values of all iterations. The average value is the variational Lyapunov index indicator of the mutation point, that is, the complexity of the mutation point.

[0045] Furthermore, the specific method of calculating the comprehensive complexity threshold using the random candidate-Youden maximization method includes:

[0046] Step D1: Obtain a history group, a known health status, and calculate the comprehensive complexity of the history group;

[0047] Step D2: randomly setting f candidate comprehensive complexity thresholds using a random method;

[0048] Step D3: sort the comprehensive complexity of the historical group in ascending order, and evenly divide the comprehensive complexity of the historical group into N intervals;

[0049] Step D4, randomly selecting f interval candidate comprehensive complexity thresholds from each interval using a random method;

[0050] Step D5: Calculate the true positive rate and false positive rate for each candidate comprehensive complexity threshold of the interval;

[0051] Step D6, calculating the Youden index based on the true positive rate and false positive rate under the same interval candidate comprehensive complexity threshold, and obtaining the Youden index of f interval candidate comprehensive complexity thresholds;

[0052] Step D7, comparing the Youden index of candidate comprehensive complexity thresholds in the same interval, in the range of [0,1];

[0053] Step D8: Select the candidate comprehensive complexity threshold of the interval with the largest Youden index as the final comprehensive complexity threshold of the interval;

[0054] Step D9: Repeat steps D4 to D8 to obtain N interval comprehensive complexity thresholds; based on the N intervals, use the quadratic fitting method to obtain the comprehensive complexity threshold.

[0055] A device health monitoring data intelligent processing system, which is applied to implement the device health monitoring data intelligent processing method, comprises:

[0056] Dynamic collection mechanism design module: Design a dynamic collection mechanism to continuously collect device status data at k moments;

[0057] Equipment status data ranking module: uses the time series ranking method to obtain the ranking vector of equipment status data at each moment;

[0058] Comprehensive impact calculation module: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, the rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment;

[0059] Data group division module: Using the comprehensive impact data at the initial moment as the benchmark, the module compares the comprehensive impact data at subsequent moments using the Euclidean distance and sets a distance threshold. When the comprehensive impact data at moment i exceeds the threshold compared to the current benchmark, the benchmark is updated and the device status data corresponding to the previous comprehensive impact data are grouped together. This operation is repeated until all device status data at all moments have been processed, and the device status data at k consecutive moments are divided into d groups.

[0060] Comprehensive complexity assessment module: For the device status data in each of the d groups, a dynamic mutation detection method is used to calculate the comprehensive complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexities of all mutation points in the group are averaged to obtain the comprehensive complexity of the group.

[0061] Health judgment module: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

[0062] The technical effects and advantages of the device health monitoring data intelligent processing method and system of the present invention are as follows:

[0063] The present invention realizes intelligent analysis of the health status of equipment by designing a dynamic collection mechanism, time series ranking, rank value mapping transformation measurement method, dynamic mutation detection method and random candidate-Youden maximization method. It can automatically identify abnormal status of equipment without human intervention and improve monitoring efficiency.

[0064] Provides basic and enhanced modes, which can dynamically adjust the type of data collected based on environmental changes (such as temperature increase), increase attention to key status data, and reduce unnecessary data storage and computing burdens;

[0065] Automatically divide continuous time series data into multiple groups, making health status assessment more detailed, able to detect different status changes in equipment operation, and improve the sensitivity of fault detection;

[0066] For each group, a variety of data processing methods (such as logarithmic transformation, exponential transformation, Euclidean distance, variational Lyapunov index, etc.) are used to calculate the comprehensive impact data and complexity indicators of the equipment status, ensuring more accurate judgment of health status and reducing misjudgments and missed judgments;

[0067] The random candidate-Youden maximization method is used to calculate the optimal comprehensive complexity threshold based on historical data to ensure optimal classification performance of healthy and unhealthy states and avoid misjudgment problems caused by fixed thresholds;

[0068] Overall, this technical solution can effectively improve the intelligence level of equipment health status monitoring, improve accuracy, sensitivity and adaptability, and has important application value for the intelligent operation and maintenance of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a schematic diagram of a method for intelligently processing equipment health monitoring data according to the present invention;

[0070] Figure 2 Schematic diagram of a method for dividing device status data at k consecutive moments into d groups according to the present invention;

[0071] Figure 3 This is a schematic diagram of an intelligent processing system for equipment health monitoring data according to the present invention. DETAILED DESCRIPTION

[0072] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0073] Example 1

[0074] See also Figure 1 and Figure 2 As shown, this embodiment provides a method for intelligently processing equipment health monitoring data, including:

[0075] Step SS1: Design a dynamic collection mechanism to collect device status data for k consecutive moments;

[0076] Step SS2: Use the time series ranking method to obtain the ranking vector of the device status data at each moment;

[0077] Step SS3: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, a rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment;

[0078] Step SS4: Using the comprehensive impact data at the initial moment as a benchmark, compare the comprehensive impact data at subsequent moments using the Euclidean distance, and set a distance threshold. When the comprehensive impact data at moment i exceeds the threshold compared to the current benchmark, the benchmark is updated, and the device status data corresponding to the previous comprehensive impact data are grouped together. Repeat this process until all device status data at all moments have been processed, and the device status data for k consecutive moments are divided into d groups.

[0079] Step SS5: Based on the device status data in each group, a dynamic mutation detection method is used to calculate the overall complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexity of all mutation points in the group is averaged to obtain the overall complexity of the group.

[0080] Step SS6: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

[0081] In existing systems, all data points are collected. Even when no additional data is needed, redundant data collection is performed, which increases the burden on the system (storage, transmission and computing costs).

[0082] Moreover, it cannot adapt to environmental changes. The fixed collection mode cannot dynamically adjust the collection strategy according to actual environmental conditions (such as temperature changes), thus failing to optimize the collection timing and data types.

[0083] Specific ways to design a dynamic collection mechanism include:

[0084] The dynamic collection mechanism includes basic mode and enhanced mode. The device status data collected in basic mode only includes key status data;

[0085] The device status data collected in enhanced mode includes key status data and auxiliary status data;

[0086] The trigger condition of the basic mode is always mandatory acquisition, which is not affected by other factors;

[0087] The trigger condition of the enhanced mode is dynamically enabled on demand. When the ambient temperature is higher than the standard ambient temperature, the enhanced mode is triggered, and the collected device status data includes key status data and auxiliary status data;

[0088] By designing basic and enhanced modes, the data collection strategy can be dynamically adjusted according to ambient temperature conditions. When the temperature is too high, the enhanced mode is enabled to collect more comprehensive data; when the temperature is normal, only key status data is collected. This on-demand dynamic activation strategy not only reduces unnecessary data collection but also improves collection efficiency.

[0089] Adjust data collection mode based on ambient temperature to ensure sufficient critical information is collected to avoid equipment damage or inaccurate health assessments caused by environmental factors;

[0090] Moreover, switching between basic mode and enhanced mode helps reduce unnecessary data storage and processing pressure, optimizes the system's computing resources and bandwidth consumption, and avoids excessive data accumulation.

[0091] Key status data include bearing vibration data, gearbox vibration data, motor vibration data, motor temperature data, winding temperature data, lubricating oil temperature data, hydraulic system pressure data, pneumatic system pressure data, cooling system pressure data, motor current data, and motor voltage data;

[0092] Among them, the vibration data of bearings, gearboxes and motors are collected through vibration sensors installed on the bearings, gearboxes and motor surfaces; the temperature data of motors, windings and lubricating oils are collected through temperature sensors installed on the motor surface, inside the motor windings and in the lubricating oil system;

[0093] The pressure data of the hydraulic system, the pressure data of the pneumatic system and the pressure data of the cooling system are collected by pressure sensors installed on the oil pump or oil pipe of the hydraulic system, the pneumatic system and the cooling system;

[0094] The motor current data is collected through the current sensor, and the motor voltage data at the motor input end is monitored through the voltage sensor;

[0095] When the ambient temperature is higher than the standard ambient temperature, the device status data includes key status data and auxiliary status data, wherein the auxiliary status data includes motor vibration acceleration data and vibration velocity data;

[0096] The motor vibration acceleration data is collected by installing an acceleration sensor on the motor bearing, and the vibration velocity data is collected by installing a velocity sensor on the motor housing.

[0097] The specific methods of using the time series ranking method to obtain the ranking vector of the device status data at each moment include:

[0098] Retrieve the device status data at k moments, where the device status data includes key status data and auxiliary status data, or only key status data. If the device status data only includes key status data, the data values of the auxiliary status data at that moment are all assumed to be 0.

[0099] Sort the data values in the nth dimension at k moments in ascending order, with the smallest data value ranked 1 and the largest data value ranked k;

[0100] Calculate the ranking of the data values of each dimension at each moment and generate a ranking vector;

[0101] Data points with the same data value are given the same rank.

[0102] Existing methods typically only consider data at the current moment and fail to capture the changing trends or time series characteristics of the device's health status. For example, a device may show an abnormal health status at a certain moment, but this abnormality may only be a temporary fluctuation. Relying only on current data may lead to misjudgment.

[0103] If only data at a single moment is considered, it will be affected by short-term fluctuations, resulting in inaccurate monitoring of device health status. Devices may experience temporary abnormalities due to factors such as environmental interference and instantaneous load changes, and these changes do not represent the long-term health of the device.

[0104] Focusing solely on data at a single moment in time cannot fully assess the evolution of a device's status over time. The evolution of a device's health status is a dynamic process, requiring comprehensive analysis based on data from multiple moments in time to more accurately capture device health trends.

[0105] Collect device status data at k moments, use the rank value mapping transformation measurement method to obtain the comprehensive impact data at each moment, take the first comprehensive impact data as the benchmark, calculate the distance between the first comprehensive impact data and the comprehensive impact data at the second, third, fourth, etc. moments using the Euclidean distance method, and set a distance threshold. If the distance between the first comprehensive impact data and the fourth comprehensive impact data is greater than the distance threshold, use the fourth similarity value as the second benchmark, and set the device status data corresponding to the comprehensive impact data from the first to the third moments as the first group;

[0106] Then calculate the distance between the comprehensive impact data at the fourth moment and the comprehensive impact data at the fifth, sixth, seventh, eighth, and ninth moments. If the distance between the comprehensive impact data at the fourth moment and the comprehensive impact data at the ninth moment is greater than the distance threshold, the comprehensive impact data at the ninth moment is used as the third benchmark; and the device status data corresponding to the comprehensive impact data from the fourth to the eighth moments is set as the second group.

[0107] Repeat the operation until the comprehensive impact data at the kth moment is calculated. In this way, the device status data at k moments are divided into d groups.

[0108] The distance threshold is calculated based on multiple experimental results;

[0109] This invention uses dynamic grouping based on time series data and determines the group division by calculating the difference between the comprehensive impact data according to the Euclidean distance. This method can automatically adjust the grouping based on the real-time comprehensive impact data of the device, ensuring that the data status within each group is similar, and more accurately reflecting the changes in the device.

[0110] Automatically updating benchmark points and groups based on distance thresholds avoids the limitations of fixed grouping methods and can flexibly respond to the influence of interference factors. This method can not only adapt to dynamic changes in equipment, but also reduce the collection of redundant data, improving system efficiency and prediction accuracy.

[0111] The specific methods of using the rank value mapping transformation measurement method to obtain the comprehensive impact data at each moment include:

[0112] The device status data at each moment includes n data points, that is, n dimensions. Each data point has a specific data value and ranking; therefore, each moment has a data value vector and ranking vector.

[0113] Using a normalization method to scale the data values in the data value vector to [0, 1], thereby obtaining scaled data values; performing a logarithmic transformation based on the scaled data values, thereby obtaining logarithmic transformed data values;

[0114] Use the normalization method to scale the ranking in the ranking vector to [0, 1] to obtain the scaled ranking; perform exponential transformation based on the scaled ranking to obtain the exponentially transformed ranking;

[0115] The influence data of a data point is obtained by performing a weighted summation of the logarithmic transformed data value and the exponential transformed ranking of the data point;

[0116] Repeat the calculation to obtain the impact data of n data points;

[0117] And use the averaging method to average the impact data of n data points to obtain the comprehensive impact data;

[0118] The weights of the logarithmically transformed data values and the exponentially transformed rankings in the weighted sum are: Among them, v j is the logarithmic transformed data value of the jth dimension, m is the total number of dimensions, wv j is the initial weight of the logarithmic transformed data value of the j-th dimension; Among them, pj is the ranking after exponential transformation of the j-th dimension, wp j is the initial weight of the ranking after the exponential transformation of the j-th dimension;

[0119] The weights of the logarithmic-transformed data values and the weights of the exponential-transformed rankings are calculated based on the initial weights of the logarithmic-transformed data values and the initial weights of the exponential-transformed rankings. The formula is: in, and are the weights of the data values after logarithmic transformation and the weights of the rankings after exponential transformation, respectively;

[0120] Assume that the device status data includes motor temperature data, pressure data, and motor vibration data. At time A, the motor temperature data is 85°C, the air pressure system pressure data is 130kPa, and the motor vibration data is 5.2mm / s^2; at time B, the motor temperature data is 92°C, the air pressure system pressure data is 135kPa, and the motor vibration data is 7.8mm / s^2; at time C, the motor temperature data is 78°C, the air pressure system pressure data is 110kPa, and the motor vibration data is 7.5mm / s^2; at time D, the motor temperature data is 88°C, the air pressure system pressure data is 125kPa, and the motor vibration data is 6.1mm / s^2.

[0121] Then the data value vector at time A is [85, 130, 5.2]; the data value vector at time B is [92, 135, 7.8]; the data value vector at time C is [78, 110, 7.5]; the data value vector at time D is [88, 125, 6.1];

[0122] Ranking of motor temperature data: The motor temperature data at time A is 85°C, ranking second; the motor temperature data at time B is 92°C, ranking fourth; the motor temperature data at time C is 78°C, ranking first; and the motor temperature data at time D is 88°C, ranking third;

[0123] The ranking of the air pressure system pressure data: the air pressure system pressure data at time A is 130kPa, ranking third; the air pressure system pressure data at time B is 135kPa, ranking fourth; the air pressure system pressure data at time C is 110kPa, ranking first; and the air pressure system pressure data at time D is 125kPa, ranking second;

[0124] Ranking of motor vibration data: The motor vibration data at time A is 5.2 mm / s^2, ranking first; the motor vibration data at time B is 7.8 mm / s^2, ranking fourth; the motor vibration data at time C is 7.5 mm / s^2, ranking third; and the motor vibration data at time D is 6.1 mm / s^2, ranking second.

[0125] Then, the ranking vector at time A is [2, 3, 1], the ranking vector at time B is [4, 4, 4], the ranking vector at time C is [1, 1, 3], and the ranking vector at time D is [3, 2, 2];

[0126] Use logarithmic transformation to calculate the motor temperature data. The logarithmic transformation formula is: V = log(1 + k * y), where k is the adjustment factor of 10, V is the motor temperature data after logarithmic transformation, and y is the normalized motor temperature data;

[0127] The ranking is calculated using the exponential decay method. The formula for the exponential decay method is: P = e -w*r , where P is the motor temperature index ranking, r is the normalized ranking, and w is the adjustment factor of 5;

[0128] At time A, a weighted summation is performed on the motor temperature data values after logarithmic transformation and the temperature ranking after exponential transformation to obtain the motor temperature impact data at time A. The air pressure system pressure impact data and the motor vibration impact data at time A are also calculated. The average method is used to calculate the comprehensive impact data of these three data points at time A.

[0129] The idea behind using data values and ranking weights to generate comprehensive impact data is to comprehensively reflect two aspects of device status: one is the actual measured value of the device (such as temperature, pressure, vibration, etc.), and the other is the relative quality of the device status in the time series (i.e., ranking). Combining these two factors provides a more comprehensive and dynamic evaluation standard for device health assessment.

[0130] The data value reflects the actual status of the device, usually the specific numerical value of the device's physical quantity (such as temperature, pressure, vibration, etc.). This data can directly describe the health status of the device. For example, excessive temperature or excessive pressure on the device may mean that the device is at risk of failure or is about to fail.

[0131] Ranking reflects the relative quality of device status, that is, the comparison between the status at a certain moment and that at other moments. Ranking can capture relative changes and help focus on the changes in device status over time.

[0132] Use logarithmic transformation on data values: Equipment status data contains some very large values (such as extreme temperature values and instantaneous pressure fluctuations). These extreme values have a disproportionate impact on weighted calculations, causing the model to favor certain data points. By performing a logarithmic transformation on the data, the differences in extreme values are compressed, thereby reducing their impact on the calculation results.

[0133] The logarithmic transformation not only compresses large values but also effectively amplifies changes in small values. It is particularly sensitive to changes within a small range. In many real-world scenarios, sudden changes or anomalies in device status often manifest as small changes. Using the logarithmic transformation can help capture these subtle changes more keenly, improving the sensitivity of anomaly detection.

[0134] Ranking itself is a linear transformation that simply identifies the relative position of each data point. However, the difference between rankings may not directly reflect the influence of the data point. Exponential transformation increases the weight of ranking differences, making the impact of ranking differences on the final result more significant.

[0135] Evaluating the data value at a certain moment alone may be affected by noise or short-term anomalies, resulting in unstable evaluation results. However, by aggregating data from multiple similar moments into a group for evaluation, accidental fluctuations can be smoothed out and the impact of noise on the final evaluation results can be reduced.

[0136] The specific method of calculating the comprehensive complexity of the group using a dynamic mutation detection method includes:

[0137] Step a1: for the dth group, the group includes device status data at h moments;

[0138] Step a2: Use the Kalman filter method to smooth the data values of the device status data at h moments to obtain the filtered values of the device status data at h moments, and calculate the residual between the filtered values of the device status data at h moments and the data values of the device status data at h moments; if the residual exceeds a dynamic threshold, the data point corresponding to the data value is a mutation point; wherein, in the Kalman filter, the dynamic threshold is usually not a predefined fixed value, but is automatically calculated by the covariance matrix during the filtering process;

[0139] Step a3: For the mutation point, use the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point;

[0140] Step a4, averaging the complexity of all mutation points in the dth group to obtain a comprehensive complexity index of the dth group;

[0141] Device status data often contains random noise and sudden fluctuations. Directly using raw data for mutation detection may lead to misjudgment. Kalman filtering can remove noise without losing important information, improving the accuracy of mutation point detection.

[0142] A single mutation point may not fully reflect the overall health status of the equipment because the mutation point may be caused by factors such as instantaneous environmental changes and load fluctuations. However, a comprehensive evaluation of all mutation points in a group can provide a more stable and comprehensive health status judgment and reduce the interference of accidental factors.

[0143] Sudden changes in data points in device status data may be due to normal fluctuations or may be a precursor to an abnormality or failure. Existing methods cannot determine the nature of the sudden change point based solely on the amplitude of data fluctuations (such as standard deviation, coefficient of variation, etc.);

[0144] For mutation points, the specific methods of using the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point include:

[0145] Step c1: When a mutation point is detected, a disturbance point is added to the data value of the mutation point to form a disturbance initial value; for the disturbance point, H disturbance points are randomly generated using a random method, and corresponding to H disturbance initial values;

[0146] For example: x'=x+ε, where ε is the disturbance point, x is the data value of the mutation point, and x' is the initial value of the disturbance; randomly set H disturbance points ε1, ε2, ε3, ..., ε H , calculate H initial disturbance values according to H disturbance points;

[0147] Step c2: For each generated initial perturbation value, compare it with the data value of the mutation point to obtain the initial error;

[0148] Such as: d h =|x' h -x|, where d h is the initial error of the hth disturbance point, x' h is the initial value of the disturbance at the h-th disturbance point;

[0149] Step c3: Set a fixed number of iterations and a linear function, and use the linear function to iteratively calculate the mutation point and its initial perturbation value to obtain the trajectory of the mutation point and the initial perturbation value;

[0150] For example: set the number of iterations and continuously iterate the linear function to obtain the trajectory of the mutation point x, x 1 , x 2 ,…,x T ; and obtain the initial value trajectory of the disturbance, x', x' 1 , x' 2 ,…,x' T ; where x 1 、x 2 ,…,x Tis the mutation point data value of the first iteration, the second iteration and the Tth iteration, T is the number of iterations; x' 1 、x' 2 ,…,x' T is the initial value of the perturbation for the first iteration, the second iteration, and the Tth iteration;

[0151] Step c4: In each iteration, the difference between the mutation point trajectory and the perturbation initial value trajectory is calculated;

[0152] like: Among them, x' h t represents the initial value of the perturbation of the hth perturbation point in the tth iteration, x t The data value representing the mutation point in the t-th iteration; represents the difference between the mutation point in the tth iteration and the initial value of the perturbation, where t is the iteration index;

[0153] Step c5: Take the natural logarithm of the relative error in each iteration to obtain a logarithmic value. The relative error is the ratio of the difference between the mutation point trajectory and the perturbation initial value trajectory in the iteration to the initial error. After the iteration is completed, use the averaging method to average the logarithmic values of all iterations to obtain the local Lyapunov exponent indicator; such as: λ is the local Lyapunov exponent index;

[0154] The calculation of the local Lyapunov exponent requires measuring the exponential growth of the relative error. Exponential growth shows a linear trend on a logarithmic scale, so taking the logarithm can linearize the error growth, making the calculation more stable.

[0155] Step c6: Using steps c1 to c5, calculate the local Lyapunov exponent index for H disturbance points; select the disturbance point corresponding to the maximum value of the local Lyapunov exponent index as the optimal disturbance point ε * ;

[0156] Step c7: Using the optimal perturbation point, calculate the mutation point trajectory and the optimal perturbation point trajectory, and calculate the logarithmic value of each iteration. After the iteration is completed, use the averaging method to average the logarithmic values of all iterations. The average value is the variational Lyapunov index indicator of the mutation point. The variational Lyapunov index indicator can reflect the complexity of the mutation point.

[0157] A mutation point typically indicates a dramatic change or turning point in the system state, reflecting a critical moment when equipment failure or performance degradation may occur. Calculating the complexity of a mutation point can reveal the system's sensitivity to small disturbances and thus identify system instability. A high complexity of a mutation point usually means that the system may be in a chaotic state or highly unstable near that point, thus helping to identify potential failure risks.

[0158] During operation, equipment may experience a variety of different states, including normal, minor anomalies, and serious faults. The complexity of the mutation point provides a method to quantify and distinguish different health states. Under normal conditions, system changes are relatively stable, while under abnormal or faulty conditions, the complexity of the mutation point is usually higher. By analyzing the complexity of the mutation point, the health status of the equipment can be accurately diagnosed, helping maintenance personnel make more precise decisions.

[0159] By assessing the complexity of mutation points, maintenance personnel can identify which mutations may cause serious problems and which are just short-term changes; for moments with higher mutation complexity, inspection and maintenance can be prioritized to reduce the probability of failure; while for mutation points with lower complexity, excessive intervention is not required, thereby reducing unnecessary maintenance costs and downtime.

[0160] Existing methods usually use empirical or statistical methods to set thresholds. Although these methods can work effectively in many cases, they often have certain limitations, mainly manifested in their reliance on subjective experience, inability to adapt to changing data, lack of flexibility and versatility, etc., which can easily lead to the set thresholds being unrepresentative.

[0161] The specific methods for calculating the comprehensive complexity threshold using the random candidate-Youden maximization method include:

[0162] Step D1: Obtain a history group, a known health status, and calculate the comprehensive complexity of the history group;

[0163] Step D2: randomly setting f candidate comprehensive complexity thresholds using a random method;

[0164] Step D3: sort the comprehensive complexity of the historical group in ascending order, and evenly divide the comprehensive complexity of the historical group into N intervals;

[0165] Step D4, randomly selecting f interval candidate comprehensive complexity thresholds from each interval using a random method;

[0166] Step D5: Calculate the true positive rate and false positive rate for each candidate comprehensive complexity threshold of the interval;

[0167] Step D6, calculating the Youden index based on the true positive rate and false positive rate under the same interval candidate comprehensive complexity threshold, and obtaining the Youden index of f interval candidate comprehensive complexity thresholds;

[0168] Step D7, comparing the Youden index of candidate comprehensive complexity thresholds in the same interval, in the range of [0,1];

[0169] Step D8: Select the candidate comprehensive complexity threshold of the interval with the largest Youden index as the final comprehensive complexity threshold of the interval;

[0170] Step D9, repeating steps D4 to D8 to obtain N interval comprehensive complexity thresholds; using the quadratic fitting method to obtain the comprehensive complexity threshold based on the N intervals;

[0171] Because the maximized Youden index means maximizing the classification performance while balancing the true positive rate and false positive rate, that is, it is most effective in distinguishing between healthy and unhealthy states;

[0172] Compared with existing methods based on fixed standards or empirical threshold setting, the random candidate-Youden maximization method can adaptively select the comprehensive complexity threshold within a wide range. By generating multiple candidate thresholds and calculating their corresponding true positive rate and false positive rate, the threshold can be flexibly adjusted according to the characteristics of the dataset, avoiding the limitations of fixed standards.

[0173] The Youden index determines the threshold by balancing the true positive rate and the false positive rate. Maximizing the Youden index means selecting a threshold that optimally balances the relationship between true positives and false positives. This method ensures that when evaluating comprehensive complexity, it captures as many true positive samples as possible while reducing incorrectly identified negative samples, providing a more reliable threshold setting.

[0174] This embodiment implements intelligent analysis of device health status by designing a dynamic collection mechanism, time series ranking, rank value mapping transformation measurement method, dynamic mutation detection method, and random candidate-Youden maximization method. It can automatically identify abnormal device status without manual intervention, thereby improving monitoring efficiency.

[0175] Provides basic and enhanced modes, which can dynamically adjust the type of data collected based on environmental changes (such as temperature increase), increase attention to key status data, and reduce unnecessary data storage and computing burdens;

[0176] Automatically divide continuous time series data into multiple groups, making health status assessment more detailed, able to detect different status changes in equipment operation, and improve the sensitivity of fault detection;

[0177] For each group, a variety of data processing methods (such as logarithmic transformation, exponential transformation, Euclidean distance, variational Lyapunov index, etc.) are used to calculate the comprehensive impact data and complexity indicators of the equipment status, ensuring more accurate judgment of health status and reducing misjudgments and missed judgments;

[0178] The random candidate-Youden maximization method is used to calculate the optimal comprehensive complexity threshold based on historical data to ensure optimal classification performance of healthy and unhealthy states and avoid misjudgment problems caused by fixed thresholds;

[0179] Overall, this technical solution can effectively improve the intelligence level of equipment health status monitoring, improve accuracy, sensitivity and adaptability, and has important application value for the intelligent operation and maintenance of industrial equipment.

[0180] Example 2

[0181] See also Figure 3 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A system for intelligent processing of equipment health monitoring data is provided, including:

[0182] Dynamic collection mechanism design module: Design a dynamic collection mechanism to continuously collect device status data at k moments;

[0183] Equipment status data ranking module: uses the time series ranking method to obtain the ranking vector of equipment status data at each moment;

[0184] Comprehensive impact calculation module: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, the rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment;

[0185] Data group division module: Using the comprehensive impact data at the initial moment as the benchmark, the module compares the comprehensive impact data at subsequent moments using the Euclidean distance and sets a distance threshold. When the comprehensive impact data at moment i exceeds the threshold compared to the current benchmark, the benchmark is updated and the device status data corresponding to the previous comprehensive impact data are grouped together. This operation is repeated until all device status data at all moments have been processed, and the device status data at k consecutive moments are divided into d groups.

[0186] Comprehensive complexity assessment module: For the device status data in each of the d groups, a dynamic mutation detection method is used to calculate the comprehensive complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexities of all mutation points in the group are averaged to obtain the comprehensive complexity of the group.

[0187] Health judgment module: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

[0188] Example 3

[0189] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned method and system for intelligent processing of equipment health monitoring data is implemented.

[0190] Since the electronic device introduced in this embodiment is an electronic device used to implement a device health monitoring data intelligent processing method and system in the embodiment of this application, based on the device health monitoring data intelligent processing method and system introduced in the embodiment of this application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as technical personnel in this field implement the electronic device used in the device health monitoring data intelligent processing method and system in the embodiment of this application, it falls within the scope of protection of this application.

[0191] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0192] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent processing of equipment health monitoring data, characterized in that: include: Step SS1: Design a dynamic collection mechanism to collect device status data for k consecutive moments; Step SS2: Use the time series ranking method to obtain the ranking vector of the device status data at each moment; Step SS3: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, a rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment; Step SS4: Taking the comprehensive impact data at the initial moment as a benchmark, compare the comprehensive impact data at subsequent moments with the comprehensive impact data through Euclidean distance, and set a distance threshold; When the comprehensive impact data at time i exceeds the threshold compared to the current benchmark, the benchmark is updated and the device status data corresponding to the previous comprehensive impact data are grouped together. This operation is repeated until all device status data at all moments are processed, and the device status data at k consecutive moments are divided into d groups. Step SS5: Based on the device status data in each group, a dynamic mutation detection method is used to calculate the overall complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexity of all mutation points in the group is averaged to obtain the overall complexity of the group. Step SS6: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

2. The method for intelligent processing of equipment health monitoring data according to claim 1, characterized in that: The specific methods of designing the dynamic acquisition mechanism include: The dynamic collection mechanism includes basic mode and enhanced mode. The device status data collected in basic mode only includes key status data; The device status data collected in enhanced mode includes key status data and auxiliary status data; The trigger condition of the basic mode is always mandatory acquisition, which is not affected by other factors; The trigger condition of the enhanced mode is dynamic activation on demand. When the ambient temperature is higher than the standard ambient temperature, the enhanced mode is triggered, and the collected device status data includes key status data and auxiliary status data.

3. The method for intelligent processing of equipment health monitoring data according to claim 2, characterized in that: The key status data includes bearing vibration data, gearbox vibration data, motor vibration data, motor temperature data, winding temperature data, lubricating oil temperature data, hydraulic system pressure data, pneumatic system pressure data, cooling system pressure data, motor current data, and motor voltage data.

4. The method for intelligent processing of equipment health monitoring data according to claim 2, characterized in that: When the ambient temperature is higher than the standard ambient temperature, the device status data includes key status data and auxiliary status data, wherein the auxiliary status data includes motor vibration acceleration data and vibration velocity data.

5. The method for intelligent processing of equipment health monitoring data according to claim 1, characterized in that: The specific method of using the time series ranking method to obtain the ranking vector of the device status data at each moment includes: Retrieve the device status data at k moments, where the device status data includes key status data and auxiliary status data, or only key status data. If the device status data only includes key status data, the data values of the auxiliary status data at that moment are all assumed to be 0. Sort the data values in the nth dimension at k moments in ascending order, with the smallest data value ranked 1 and the largest data value ranked k; Calculate the ranking of the data values of each dimension at each moment and generate a ranking vector; Data points with the same data value are given the same rank.

6. The method for intelligent processing of equipment health monitoring data according to claim 5, characterized in that: The specific method of using the rank value mapping transformation measurement method to obtain the comprehensive impact data at each moment includes: The device status data at each moment includes n data points, that is, n dimensions. Each data point has a specific data value and ranking; therefore, each moment has a data value vector and ranking vector. Using a normalization method to scale the data values in the data value vector to [0, 1], thereby obtaining scaled data values; performing a logarithmic transformation based on the scaled data values, thereby obtaining logarithmic transformed data values; Use the normalization method to scale the ranking in the ranking vector to [0, 1] to obtain the scaled ranking; perform exponential transformation based on the scaled ranking to obtain the exponentially transformed ranking; The influence data of a data point is obtained by performing a weighted summation of the logarithmic transformed data value and the exponential transformed ranking of the data point; Repeat the calculation to obtain the impact data of n data points; The average method is used to average the impact data of n data points to obtain the comprehensive impact data.

7. The method for intelligent processing of equipment health monitoring data according to claim 1, characterized in that: The specific method of calculating the comprehensive complexity of the group using a dynamic mutation detection method includes: Step a1: for the dth group, the group includes device status data at h moments; Step a2: Use the Kalman filter method to smooth the data values of the device status data at h moments to obtain the filtered values of the device status data at h moments, and calculate the residual between the filtered values of the device status data at h moments and the data values of the device status data at h moments; if the residual exceeds the dynamic threshold, the data point corresponding to the data value is a mutation point; Step a3: For the mutation point, use the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point; Step a4: averaging the complexity of all mutation points in the dth group to obtain the comprehensive complexity index of the dth group.

8. The method for intelligent processing of equipment health monitoring data according to claim 7, characterized in that: The specific method of using the perturbation method combined with the variational Lyapunov exponent calculation method to evaluate the complexity of the mutation point includes: Step c1: When a mutation point is detected, a disturbance point is added to the data value of the mutation point to form a disturbance initial value; for the disturbance point, H disturbance points are randomly generated using a random method, and corresponding to H disturbance initial values; Step c2: For each generated initial perturbation value, compare it with the data value of the mutation point to obtain the initial error; Step c3: Set the number of iterations and the linear function, and use the linear function to iteratively calculate the mutation point and its initial perturbation value to obtain the trajectory of the mutation point and the initial perturbation value; Step c4: In each iteration, the difference between the mutation point trajectory and the perturbation initial value trajectory is calculated; Step c5: Take the natural logarithm of the relative error in each iteration to obtain a logarithm value. The relative error is the ratio of the difference between the mutation point trajectory and the perturbation initial value trajectory in the iteration to the initial error. After the iteration is completed, use the averaging method to calculate the average of the logarithm values of all iterations, which is the local Lyapunov exponent indicator. Step c6: Using steps c1 to c5, calculate the local Lyapunov exponent index for H disturbance points; select the disturbance point corresponding to the maximum value of the local Lyapunov exponent index as the optimal disturbance point; Step c7: Using the optimal perturbation point, calculate the mutation point trajectory and the optimal perturbation point trajectory, and calculate the logarithmic value of each iteration. After the iteration is completed, use the averaging method to calculate the average of the logarithmic values of all iterations. The average value is the variational Lyapunov index indicator of the mutation point, that is, the complexity of the mutation point.

9. The method for intelligent processing of equipment health monitoring data according to claim 8, characterized in that: The specific method of calculating the comprehensive complexity threshold using the random candidate-Youden maximization method includes: Step D1: Obtain a history group, a known health status, and calculate the comprehensive complexity of the history group; Step D2: randomly setting f candidate comprehensive complexity thresholds using a random method; Step D3: sort the comprehensive complexity of the historical group in ascending order, and evenly divide the comprehensive complexity of the historical group into N intervals; Step D4, randomly selecting f interval candidate comprehensive complexity thresholds from each interval using a random method; Step D5: Calculate the true positive rate and false positive rate for each candidate comprehensive complexity threshold of the interval; Step D6, calculating the Youden index based on the true positive rate and false positive rate under the same interval candidate comprehensive complexity threshold, and obtaining the Youden index of f interval candidate comprehensive complexity thresholds; Step D7, comparing the Youden index of candidate comprehensive complexity thresholds in the same interval; Step D8: Select the candidate comprehensive complexity threshold of the interval with the largest Youden index as the final comprehensive complexity threshold of the interval; Step D9: Repeat steps D4 to D8 to obtain N interval comprehensive complexity thresholds; based on the N intervals, use the quadratic fitting method to obtain the comprehensive complexity threshold.

10. An intelligent processing system for equipment health monitoring data, applied to an intelligent processing method for equipment health monitoring data according to any one of claims 1 to 9, characterized in that: include: Dynamic collection mechanism design module: Design a dynamic collection mechanism to continuously collect device status data at k moments; Equipment status data ranking module: uses the time series ranking method to obtain the ranking vector of equipment status data at each moment; Comprehensive impact calculation module: Based on the ranking vector of the device status data at each moment and the data value vector of the device status data at each moment, the rank value mapping transformation measurement method is used to obtain the comprehensive impact data at each moment; Data group division module: Taking the comprehensive impact data at the initial moment as the benchmark, it compares the comprehensive impact data at subsequent moments through Euclidean distance and sets a distance threshold; When the comprehensive impact data at time i exceeds the threshold compared to the current benchmark, the benchmark is updated and the device status data corresponding to the previous comprehensive impact data are grouped together. This operation is repeated until all device status data at all moments are processed, and the device status data at k consecutive moments are divided into d groups. Comprehensive complexity assessment module: For the device status data in each of the d groups, a dynamic mutation detection method is used to calculate the comprehensive complexity of the group. The Kalman filter method is used to identify the mutation points. The perturbation method combined with the variational Lyapunov exponent calculation method is used to evaluate the complexity of the mutation points. The complexities of all mutation points in the group are averaged to obtain the comprehensive complexity of the group. Health judgment module: Use the random candidate-Youden maximization method to calculate the comprehensive complexity threshold. If the comprehensive complexity of the group is less than the comprehensive complexity threshold, the devices at the moment included in the group are in a healthy state; if the comprehensive complexity of the group is greater than or equal to the comprehensive complexity threshold, the devices at the moment included in the group are in an unhealthy state.

Images