A complex equipment health assessment method based on an improved martens system
By improving the Martin system and the directed acyclic graph structure, and combining orthogonal experiments to screen characteristic parameters, the problems of accuracy and efficiency in health status assessment of complex equipment were solved, and efficient health status assessment and fault early warning were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2023-09-28
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies are insufficient to accurately assess the health status of complex equipment, and traditional maintenance methods cannot meet the needs of modern equipment maintenance and support.
A health assessment method based on an improved Martin system was adopted. Multiple MTS classifiers were constructed through data processing, state partitioning, discrimination matrix and directed acyclic graph structure. Orthogonal experiments and signal-to-noise ratio were used to screen feature parameters, and threshold sliding was set to assess health status.
It improves the accuracy and reliability of health status assessment for complex equipment, simplifies the classification process, reduces computational complexity and time, and provides a reference for real-time fault warning and maintenance.
Smart Images
Figure CN117556293B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of health status assessment of complex equipment, specifically a method for assessing the health status of complex equipment. Background Technology
[0002] The foundation for the normal operation of complex equipment lies in its maintenance and upkeep. With the increasing integration and informatization of equipment systems, the difficulty of fault diagnosis and logistical support is constantly growing. Traditional maintenance methods such as reactive and scheduled maintenance have many drawbacks and can no longer meet the needs of modern equipment maintenance and support.
[0003] Currently, condition-based maintenance (PHM) for complex equipment has become a trend. PHM technology has been extensively researched and applied in military powers such as the UK and the US, and is an important component of large and complex equipment. Health status assessment is one of the key technologies of PHM, and accurately assessing the health status of equipment is the primary basis for condition-based maintenance, which is of great significance.
[0004] To address the aforementioned issues, a health assessment method for complex equipment based on an improved Martin system is proposed. Summary of the Invention
[0005] To improve the general assessment of the health status of complex equipment using traditional health assessment methods, this invention provides an effective health status assessment method for complex equipment based on an improved Martin system.
[0006] The technical solution adopted by this invention to solve its technical problem is:
[0007] A method for health assessment of complex equipment based on an improved Martin system includes the following steps:
[0008] S1: Standardize the data from various sensors in complex equipment to eliminate the influence between different units of measurement;
[0009] S2: Divide the complete degradation process of complex equipment into three states;
[0010] S3: Construct a discrimination matrix and use the structure of a directed acyclic graph to construct a graph structure for judging the health status of complex equipment systems. Construct multiple MTS classifiers based on the graph structure.
[0011] S4: Based on different MTS classifiers, apply the robustness principle of the Martin system to the sensor parameters, select useful feature parameters to build different feature combinations, and construct sub-MTS classifiers for different feature combinations.
[0012] S5: Set threshold sliding to determine the threshold of the MTS classifier. For each MTS classifier, the classification results of the sub-MTS classifiers constructed using different feature combinations are classified based on the "majority rule" criterion.
[0013] S6: Collect sensor signals from complex equipment systems in unknown health states, and determine their health status through the classification process in S5, thereby achieving a health status assessment of complex equipment systems.
[0014] A complex equipment health assessment system based on an improved Martin system includes:
[0015] The data processing module is used to perform data normalization, noise filtering, and dimensionality reduction processing on the complex sensor data of various equipment collected by the sensor module.
[0016] The state division module is used to divide the complete degradation process of complex equipment into three states.
[0017] The discrimination calculation module is used to calculate the discrimination matrix between different health status categories;
[0018] The classification process construction module is used to construct a directed acyclic graph structure using the discrimination matrix, determine the classification process of complex equipment health status, and build an MTS classifier on each non-leaf node.
[0019] The feature selection module is used to select effective feature parameters for each MTS classifier using orthogonal experiments and signal-to-noise ratio, and to construct multiple sub-MTS classifiers based on different feature combinations.
[0020] The threshold sliding module is used to set the threshold sliding for each MTS classifier, determine its optimal threshold, and vote on the classification results of sub-MTS classifiers constructed with different feature combinations to obtain the final classification result.
[0021] The health assessment module is used to collect sensor signals from complex equipment systems in unknown health states. Through the classification process of the threshold sliding module, its health status is judged, thereby realizing the health status assessment of complex equipment systems.
[0022] A storage medium storing a program for performing a complex equipment health assessment method based on an improved Martin system.
[0023] Beneficial effects
[0024] This invention utilizes Mahalanobis distance and a discrimination matrix to effectively distinguish different health status categories of complex equipment, thereby improving the accuracy and reliability of classification.
[0025] This invention utilizes a directed acyclic graph structure to simplify the classification process of the health status of complex equipment, reducing computational complexity and time.
[0026] This invention utilizes orthogonal experiments and signal-to-noise ratio to screen out feature parameters that contribute to classification, thereby reducing feature dimensionality and redundancy.
[0027] This invention utilizes threshold sliding and voting mechanisms to determine the optimal MTS classifier threshold and classification criteria, thereby improving the stability and robustness of classification.
[0028] Through the above-mentioned effects, this invention utilizes the ability to collect sensor signals and assess the health status of complex equipment systems in unknown health states in real time, providing an effective reference for fault early warning and maintenance of complex equipment. Attached Figure Description
[0029] Figure 1 This is a flowchart of the complex equipment health assessment based on the improved Martin system of the present invention;
[0030] Figure 2 This refers to the sensor parameter degradation process of complex equipment based on normalization and filtering.
[0031] Figure 3 A flowchart for a complex equipment health status three-class classification based on a discrimination matrix and a directed acyclic graph;
[0032] Figure 4 This is a graph showing the classification accuracy as a function of the threshold determined by the sliding method. Detailed Implementation
[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0034] Reference Figures 1 to 4 A complex equipment health assessment technology based on an improved Martin system includes the following steps:
[0035] Step 1: Collect time-series parameters of various sensors for complex equipment from normal operation to failure.
[0036] Step Two: Perform data normalization on the various sensor data of the complex equipment obtained in Step One to eliminate differences between different units of measurement. Simultaneously, use mean filtering to filter sensor noise. The degradation trends of various sensors in the complex equipment after normalization and filtering are as follows: Figure 2 As shown in Table 1, the data was dimensionality reduced using Principal Component Analysis (PCA). Based on experience, PCA-reduced data should retain at least 95% of the information from the original data. This reduced the dimensions of various sensor data from complex equipment to 8 dimensions, resulting in a total information content of 95.6% for the reduced features. Specific examples are shown in Table 1.
[0037] Table 1
[0038] feature Information ratio feature Information ratio Feature 1 0.6937 Feature 5 0.0118 Feature 2 0.1896 Feature 6 0.0101 Feature 3 0.0192 Feature 7 0.0098 Feature 4 0.0133 Feature 8 0.0098
[0039] Step 3: Using the "bathtub curve" theory employed in reliability analysis, the complete degradation process of complex equipment is divided into three states. The number of cycles from 0% to 70% is set as "slow degradation label_0", the number of cycles from 70% to 90% is set as "rapid degradation label_1", and the number of cycles from 90% to 100% is set as "failure state label_2".
[0040] Step 4: Construct a discrimination matrix and use the structure of a directed acyclic graph to build a graph structure for judging the health status of complex equipment systems. Based on the graph structure, construct multiple MTS classifiers. After classifying the health status of complex equipment degradation data, the discrimination matrix can be obtained, thereby establishing a flowchart for health status classification.
[0041] For any two health status categories in the training data, one category is designated as the normal sample set and the other as the abnormal sample set. The Mahalanobis distance and the discrimination index are calculated, resulting in the discrimination index matrix as follows:
[0042] As shown in Table 2:
[0043]
[0044] Table 2
[0045] Three MTS classifiers were constructed based on the discrimination matrix, as shown in Table 3:
[0046] Table 3
[0047] MTS classifier normal samples Abnormal samples Distinguish categories MTS1 label_3 label_1 "Non-0" or "Non-2" MTS2 label_3 label_2 "Not 1" or "Not 2" MTS3 label_2 label_1 "Non-0" or "Non-1"
[0048] A health status classification model based on a directed acyclic graph is constructed based on the discrimination matrix, as follows: Figure 3 As shown.
[0049] In some disclosures, step four involves constructing a discrimination matrix and using a directed acyclic graph (DAG) to build a graph structure for judging the health status of complex equipment systems. Multiple MTS classifiers are then constructed based on this graph structure, including the following steps:
[0050] Step 4.1: Define the discrimination matrix. The discrimination matrix is defined as follows:
[0051] Suppose that there are n categories of health status for complex equipment, denoted as C = {C1, C2, ..., C...} n}, Class C i The feature samples of (i = 1, 2, ..., n) are a set of p-dimensional vectors X i , where p represents the monitoring parameters of the equipment's sensors.
[0052] Let C be any two different health status categories. i and C j The overall mean Mahalanobis distance is μ ij Then C i and C j Category Discrimination D ij for:
[0053]
[0054] Where, μ ij The calculation is as follows:
[0055]
[0056] Where, d ik C i Mahalanobis distance of the k-th sample in the health status category; d jt C i Mahalanobis distance of the t-th sample in the health status category; n i C i Sample size in the health status category, n j C j Sample size in the health status category.
[0057] Step 4.2: Calculate the discrimination matrix. Using μ defined in Step 4.1... ik Calculate C i and C k The discrimination index is used to obtain the discrimination matrix D.
[0058] Step 4.3: Obtain the classification process for the health status of complex equipment. Using the discrimination matrix calculated in Step 4.2, and based on the characteristics of the directed acyclic graph structure, the binary classification method based on the Martin system is extended to construct a tri-classification model. A Martin system classifier is built at each node (except leaf nodes) for multi-class classification.
[0059] Step 4.4: Determine the classification order based on the degree of differentiation between different samples. Select the largest D based on D. ij , denoted as D lp , will C l and C p Two types of samples were used to construct the MTS, representing normal and abnormal samples respectively, and placed at the root node; and C was selected. l and C p The sample sets with the highest discrimination are used to construct Martin systems as the second-layer node classifiers; similarly, the next-layer node classifiers are constructed until all categories are completely distinguished.
[0060] Step 5: Based on different MTS classifiers, apply the robustness principle of the Martin system to the sensor parameters, select useful feature parameters to build different feature combinations, and construct sub-MTS classifiers for different feature combinations.
[0061] Based on the principle of orthogonal experiments, design the orthogonal array L9(2 8 Based on the signal-to-noise ratio (SNR), features with a gain greater than 0 are retained. Feature filtering is performed on MTS1, MTS2, and MTS3 respectively, and the filtered features are shown in Table 4.
[0062] Table 4
[0063] Classifier Retained features Mean Mahalanobis distance of normal samples Mean Mahalanobis distance of outlier samples MTS1 Features 1, 4, 5, 7 0.999 12.027 MTS2 Features 1, 4, 5, 7 0.999 3.7688 MTS3 Features 1, 4, 5, 7 0.999 3.3968
[0064] In some disclosures, step five involves applying the robustness principle of the Martin system to the sensor parameters based on different MTS classifiers, selecting useful feature parameters to construct different feature combinations, and building sub-MTS classifiers for different feature combinations, including the following steps:
[0065] Step 5.1: For each non-leaf node of the MTS system in the directed acyclic graph structure, select effective features using orthogonal experiments and signal-to-noise ratio (SNR). Design an orthogonal array based on the number of features, construct a feature space based on the selected features in each row (one experiment), and calculate the Mahalanobis distance of outlier samples. Calculate the SNR value for each experiment using the following formula:
[0066]
[0067] Where m is the number of abnormal samples.
[0068] Then, the characteristic x is calculated using the following formula. j SNR increment:
[0069]
[0070] In the formula, and x j The mean SNR of participants in the experiment and the mean SNR of non-participants, when Gain j If the value is greater than 0, it means that the feature can be used to reflect the difference between samples and should be retained; otherwise, the feature should be deleted.
[0071] Step 5.2: Using the features selected in Step 5.1, construct sub-MTS classifiers based on different feature combinations. Taking MTS1, which distinguishes between "non-label_0" and "non-label_2", as an example, select m preferred features based on the SNR increment of each feature. Construct different preferred feature combinations based on these m preferred features. For different feature combinations... i (i = 1, 2, ..., 2) m-1 -1) Construct l i Individual MTS classifiers.
[0072] Step 6: Set threshold sliding and determine the threshold of the MTS classifier. For each MTS classifier, the classification results of the sub-MTS classifiers constructed using different feature combinations are classified based on the "majority rule" criterion.
[0073] The mean Mahalanobis distance for normal samples is around 1, while the mean Mahalanobis distance for abnormal samples is greater than 1. An optimal threshold is determined by accumulating threshold values. Specifically, the range between the mean Mahalanobis distances for normal and abnormal samples is defined. Within this range, a threshold is set starting from the mean Mahalanobis distance of normal samples and increasing by an equal amount. If the Mahalanobis distance of normal samples is below the threshold or the Mahalanobis distance of abnormal samples is above the threshold, the classification is correct; otherwise, it is incorrect. The threshold that maximizes the accuracy is selected as the final value.
[0074] Taking the MTS1 classifier as an example, the mean Mahalanobis distance for normal samples is 0.99949, and the mean Mahalanobis distance for abnormal samples is 12.027. The trend of classification accuracy with threshold value is as follows: Figure 4 As shown.
[0075] The results of building the MTS2 sub-classifier based on steps five and six are shown in Table 5:
[0076] Table 5
[0077] Subclassifier Feature combination threshold accuracy MTS2_1 Feature 1 2.310 92.37％ MTS2_2 Features 1 and 4 1.690 89.72％ …… …… …… …… MTS2_7 Features 1, 4, 5, 7 1.350 86.89％
[0078] Similarly, sub-classifiers for MT1 and MTS3 are constructed respectively, and the final classification criterion is determined based on the classification results of the sub-classifiers.
[0079] In some disclosures, step six involves setting a threshold sliding mechanism to determine the threshold for the MTS classifier. For each MTS classifier, the classification results of the sub-MTS classifiers constructed using different feature combinations are classified based on the "majority rule" criterion, including the following steps:
[0080] Step 6.1: Determine the method for calculating the MTS classifier threshold. Specifically, determine the range between the mean Mahalanobis distance of normal samples and the mean Mahalanobis distance of abnormal samples. Within this range, set a threshold starting with the mean Mahalanobis distance of normal samples and increasing it by an equal amount. If the Mahalanobis distance of normal samples is below the threshold and the Mahalanobis distance of abnormal samples is above the threshold, the classification is correct; otherwise, it is incorrect. Select the threshold that maximizes the accuracy as the final value.
[0081] Step 6.2: Based on the method for calculating the MTS classifier threshold determined in Step 6.1, apply the method to the l constructed in Step 5.2. i Each sub-classifier determines its threshold and classification criteria because step 5.2 constructs a total of 2 MTS sub-classifiers for the two classes of samples. m-1 If there are -1 subclasses, i.e. an odd number of subclasses, then if more than half of the subclasses' classification results support the test sample as "non-label_0", then the test sample is non-label_0; otherwise, the test sample is non-label_2.
[0082] Step 6.3: Based on step 6.2, determine the threshold and classification criteria for the sub-classifiers in the MTS classifier of each non-leaf node in the directed acyclic graph.
[0083] Step 7: Collect sensor signals from complex equipment systems in unknown health states. Through the classification process in Step 6, determine their health status, thereby achieving a health status assessment of the complex equipment system.
[0084] This invention discloses a method for health assessment of complex equipment based on an improved Martin system. The method assesses the health status of complex equipment using an improved Martin system and a directed acyclic graph (DAG) structure. First, the Martin system is used to calculate Mahalanobis distances between samples of different health states, analyzing and comparing the discriminative power of each sample combination. Then, a multi-classification process for equipment health status is constructed using a DAG. Next, features selected through orthogonal experiments are used to construct different feature combinations, and a separate Martin system is built for each feature combination. Finally, a voting process is performed to obtain the improved Martin system that ultimately classifies the equipment into different health state categories. This method improves the accuracy and speed of health status assessment for complex equipment, providing a strong guarantee for the stable and efficient operation of complex equipment.
[0085] The technical concept of this invention is as follows: This invention specifically relates to health assessment technology for complex equipment. To overcome the limitations of the binary classification model of the MTS system, a directed acyclic graph structure is incorporated to extend it into a multi-classification model. A discriminant matrix is used to construct the multi-classification process. Furthermore, to address the issue of poor classification accuracy, an improved Martin system is proposed. This involves constructing feature combinations using optimal features selected through orthogonal experiments and signal-to-noise ratio analysis. These feature combinations are then used to construct different MTS sub-classifiers. By combining the classification results of multiple MTS sub-classifiers, the final health status classification result is optimized. This enables effective health assessment of complex systems.
[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for health assessment of complex equipment based on an improved Martin system, characterized in that, Includes the following steps: S1: Process various sensor data from complex equipment to eliminate the influence between different units of measurement; S2: Divide the complete degradation process of complex equipment into three states; S3: Construct a discrimination matrix and use the structure of a directed acyclic graph to construct a graph structure for judging the health status of complex equipment systems. Construct multiple MTS classifiers based on the graph structure. S4: Based on different MTS classifiers, apply the robustness principle of the Martin system to the sensor parameters, select useful feature parameters to build different feature combinations, and construct sub-MTS classifiers for different feature combinations. S5: Set threshold sliding to determine the threshold of the MTS classifier. For each MTS classifier, the classification results of the sub-MTS classifiers constructed using different feature combinations are classified based on the "majority rule" criterion. S6: Collect sensor signals from complex equipment systems in unknown health states, and determine their health status through the classification process in S5, thereby achieving health status assessment of complex equipment systems. S3 includes the following steps: S31: Define a discrimination matrix and use Mahalanobis distance to calculate the similarity between different health status categories. The higher the discrimination, the less similar the categories are. S32: Calculate the discrimination matrix, using the formula defined in S31 to calculate the discrimination between different health status categories; S33: Obtain the classification process of complex equipment health status. Using the discrimination matrix calculated in S32 and combined with the directed acyclic graph structure, determine the classification process of complex equipment health status and establish an MTS classifier on each non-leaf node. S34: Determine the classification order based on the degree of differentiation between different samples, select the two classes of samples with the highest differentiation as normal samples and abnormal samples to construct the MTS, and place it in the root node; Select the sample set with the highest discriminative power from the root node sample to construct MTS, and use it as the second-layer node classifier; similarly, construct the next-layer node classifier until all categories are completely distinguished.
2. The method for health assessment of complex equipment based on the improved Martin system according to claim 1, characterized in that: The specific steps of S1 are as follows: S11: Collect time-series parameters of various sensors from normal operation to failure of complex equipment; S12: Perform data normalization on the various sensor data of the complex equipment obtained in S11, and use mean filtering and Kalman filtering algorithms to filter sensor noise, and use principal component analysis to reduce the dimensionality of sensor parameters.
3. The method for health assessment of complex equipment based on the improved Martin system according to claim 1, characterized in that: The specific steps of S2 are as follows: The complete degradation process of complex equipment is divided into three states: 0%–70% of the cycle count is set as "slow degradation", 70%–90% of the cycle count is set as "rapid degradation", and 90%–100% of the cycle count is set as "failure state".
4. The method for health assessment of complex equipment based on the improved Martin system according to claim 1, characterized in that: Step S4 utilizes the principle of orthogonal experimentation to design an orthogonal array, and combines the signal-to-noise ratio to retain features with feature gain greater than 0.
5. The method for health assessment of complex equipment based on the improved Martin system according to claim 4, characterized in that, Step S5 selects the threshold that maximizes the accuracy as the final value.
6. The method for health assessment of complex equipment based on the improved Martin system according to claim 1, characterized in that: S4 includes the following steps: S41: For each non-leaf node's MTS classifier, design an orthogonal array, constructing a new array based on the selected feature parameters for each row. Construct a feature space, calculate the Mahalanobis distance of outlier samples in the feature space, and use the signal-to-noise ratio as an evaluation index to screen out effective feature parameters. S42: Using the effective feature parameters selected in S41, construct different feature combinations and build a sub-MTS classifier for each feature combination.
7. The method for health assessment of complex equipment based on the improved Martin system according to claim 1, characterized in that: S5 includes the following steps: S51: Determine the interval between the mean Mahalanobis distance of normal samples and the mean Mahalanobis distance of abnormal samples. Within this interval, set a threshold starting from the mean Mahalanobis distance of normal samples and increasing it by an equal amount. Calculate the classification accuracy at each threshold. S52: Select the threshold that maximizes the accuracy as the optimal threshold for the MTS classifier, and determine the classification result of the unknown sample in the sub-MTS classifier constructed by different feature combinations based on the threshold. S53: For each MTS classifier, the "majority rule" criterion is used to vote on the classification results of the sub-MTS classifiers constructed by different feature combinations to obtain the final classification result.
8. A health assessment system for complex equipment based on an improved Martin system, characterized in that, include: The data processing module is used to perform data normalization, noise filtering, and dimensionality reduction processing on the complex sensor data of various equipment collected by the sensor module. The state division module is used to divide the complete degradation process of complex equipment into three states. The discrimination calculation module is used to calculate the discrimination matrix between different health status categories; The classification process construction module is used to construct a directed acyclic graph structure using the discrimination matrix, determine the classification process of complex equipment health status, and build an MTS classifier on each non-leaf node. The feature selection module is used to select effective feature parameters for each MTS classifier using orthogonal experiments and signal-to-noise ratio, and to construct multiple sub-MTS classifiers based on different feature combinations. The threshold sliding module is used to set the threshold sliding for each MTS classifier, determine its optimal threshold, and vote on the classification results of sub-MTS classifiers constructed with different feature combinations to obtain the final classification result. The health assessment module is used to collect sensor signals from complex equipment systems in unknown health states. Through the classification process of the threshold sliding module, its health status is judged, thereby realizing the health status assessment of complex equipment systems. The classification process construction module execution steps include: S31: Define a discrimination matrix and use Mahalanobis distance to calculate the similarity between different health status categories. The higher the discrimination, the less similar the categories are. S32: Calculate the discrimination matrix, using the formula defined in S31 to calculate the discrimination between different health status categories; S33: Obtain the classification process of complex equipment health status. Using the discrimination matrix calculated in S32 and combined with the directed acyclic graph structure, determine the classification process of complex equipment health status and establish an MTS classifier on each non-leaf node. S34: Determine the classification order based on the degree of differentiation between different samples, select the two classes of samples with the highest differentiation as normal samples and abnormal samples to construct MTS, and place them in the root node; select the sample set with the highest differentiation from the root node sample to construct MTS respectively, and use them as the second layer node classifier; similarly, construct the next layer node classifier until all categories are completely separated.
9. A storage medium, characterized in that, The storage medium contains a program for performing a complex equipment health assessment method based on an improved Martin system as described in any one of claims 1-7.