A Fault Diagnosis Method for Heavy Engineering Equipment that Integrates Temporal Pattern Recognition and Unsupervised Clustering

Abstract

This invention discloses a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, comprising: Step S1: extracting fault features based on source regression to obtain a feature matrix; Step S2: processing the feature matrix using a fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance to obtain the minimum target partition and partition matrix; Step S3: constructing a fault diagnosis model, training the fault diagnosis model, and identifying fault categories. This invention can effectively improve the accuracy of anomaly detection in sensor time series data and fault diagnosis of large engineering equipment, thereby ensuring safe equipment operation, improving construction efficiency, and reducing maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis of heavy engineering machinery, and in particular to a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering. Background Technology

[0002] In the fields of artificial intelligence and fault diagnosis of heavy construction machinery, unsupervised clustering machine learning methods and time series analysis techniques are key foundational technologies for processing sensor time series data. Their performance is crucial for ensuring safe equipment operation, improving construction efficiency, and reducing maintenance costs. Heavy construction machinery continuously generates multi-source sensor signals for tunneling parameters such as displacement and velocity during operation. However, these signals are characterized by nonlinearity, time-varying nature, and strong noise, and are easily affected by complex working conditions, sensor errors, and outliers, resulting in significant fluctuations, complex patterns, and concealed fault symptoms over time. This makes the faults of heavy construction machinery exhibit strong randomness, concealment, and complexity, thus posing significant challenges to fault diagnosis and early warning.

[0003] Traditionally, fault diagnosis for large equipment such as heavy construction machinery has mainly relied on empirical knowledge-based methods and mechanistic model-based methods. Although these technologies are widely used, they often exhibit problems such as insufficient fault feature extraction, inaccurate measurement of global change patterns, and insufficient robustness when dealing with nonlinear, noisy, and outlier sensor signals. In particular, many methods still use measurement methods based on inter-point distance or Euclidean distance, which are difficult to accurately reflect the true similarity of time series in terms of fluctuation direction, fluctuation amplitude, and overall shape.

[0004] In recent years, with the rapid development of data-driven pattern recognition and deep learning, time-varying noise processing technology, and similarity matching technology, these technologies have been widely applied in various fields due to their high efficiency and high accuracy, thus providing new ideas for fault diagnosis of heavy engineering machinery. Several published patents have been applied to fault diagnosis in recent years, such as patent application number CN121978445A, a fault diagnosis method and system for vehicle-mounted CCS integrated busbars based on multi-source data fusion; TWI916050B, a vehicle fault diagnosis method and system based on environmental self-adjusting Bayesian networks; CN111596210A, a fault diagnosis method for linear motor stator slot wedges based on statistical analysis; and CN121959184A, a fault diagnosis method for dual-memory enhanced subdomain alignment based on analog-digital fusion. Therefore, researching the use of digital intelligence technology for fault diagnosis of heavy engineering machinery is of great significance. Based on the above patents, these technologies have improved diagnostic accuracy and automation levels to a certain extent. However, there are still some limitations: real fault samples of heavy engineering machinery are scarce, which limits model training and generalization; conventional distance metrics cannot accurately measure the real similarity between signals containing noise and outliers, and it is difficult to stably characterize the overall trend of fault sequences; the accuracy and robustness of pattern recognition for high-frequency and strong noise signals are low, and the robustness of the model is still insufficient.

[0005] Therefore, how to provide a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, which can effectively improve the accuracy of anomaly detection in sensor time series data and fault diagnosis of large engineering equipment, thereby ensuring safe operation of equipment, improving construction efficiency and reducing maintenance costs.

[0007] The present invention solves the technical problem by adopting the following technical solution:

[0008] A fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering includes the following steps:

[0009] Step S1: Based on fault feature extraction derived from regression, obtain the feature matrix;

[0010] Step S2: The feature matrix is ​​processed by the fuzzy C-means unsupervised clustering method based on the morphological fluctuation consistency offset distance to obtain the minimum target partition and the partition matrix;

[0011] Step S3: Build a fault diagnosis model, train the fault diagnosis model and identify fault categories.

[0012] Furthermore, in step S1, the method for extracting fault features based on source regression is as follows: extracting fault features of heavy engineering equipment using a source regression model, performing linear fitting, and constructing feature matrices under different fault modes.

[0013] Furthermore, in step S2, the objective function of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is expressed as follows: In the formula This indicates the dimension of the sample data; Indicates sample Compared to the first Membership degree of a class; For the Nth sample data One sample; Indicates the first The center of each category; divide N data samples into c classes; express and The morphological fluctuation consistency offset distance value.

[0014] Furthermore, in step S2, the process of obtaining the minimum target partition and the partition matrix employs an iterative optimization algorithm. The entire optimization process is as follows:

[0015] Step S21, Initialize the partition matrix: Let ,in ;

[0016] Step S22, calculate the class center of each class, using the following formula: ;

[0017] Step S23, calculate the partition matrix The calculation formula is as follows: ;

[0018] Step S24: Iterate through steps S21 to S23, that is, after completing one step, let k = k + 1 and return to step S22; the stopping condition for iteration is... , A threshold, given by the user, represents the allowable error range; after the iteration stops, the minimum target partition is obtained. and partition matrix .

[0019] Furthermore, in step S3, the methods for training the fault diagnosis model and identifying fault categories include:

[0020] In the constructed fault diagnosis model, the input of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is the feature matrix extracted from the regression model; the extracted feature matrix is ​​the coefficient matrix calculated using the regression model under various system fault states. In the formula, This represents the coefficient vector extracted under each state;

[0021] coefficient matrix The data is input into a fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance. Through iterative optimization, the optimal parameters for this method are determined, and model training is completed. After training the diagnostic model, the autoregressive coefficients derived from the regression model are extracted from the test sample data. ;make ,Will The input is fed into the trained diagnostic model to obtain... The corresponding membership matrix is ​​used to determine the fault state type of the system to which the test sample belongs, thus realizing the classification and identification of the test sample.

[0022] The present invention discloses a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, which has the following beneficial effects:

[0023] (1) Improve fault diagnosis accuracy: By extracting fault features, modeling change trends and performing global similarity analysis on time series signals, this invention can more accurately identify abnormal patterns in nonlinear and noisy environments. Compared with methods that directly identify based on the original signal or traditional distance metrics, it can more effectively distinguish between normal states and different fault states, reduce misdiagnosis and missed diagnosis, and improve the accuracy, robustness and generalization ability of fault diagnosis results.

[0024] (2) Improve real-time response capability: Status identification can be completed without a large number of manually labeled samples. It can quickly process the continuously collected multi-source sensor signals. Under the conditions of rapid changes in the operating status of engineering machinery and complex fluctuations in working conditions, it can promptly detect potential anomalies and quickly provide diagnostic results, thereby shortening the fault detection time and reducing downtime risk and maintenance costs.

[0025] (3) Improve system stability: It can maintain good recognition performance under conditions of high noise, outlier interference and limited fault samples, and has strong adaptability to sensor time series data under complex working conditions, which helps to improve the reliability and stability of heavy engineering machinery monitoring system operation.

[0026] (4) Enhance decision support capabilities: With its ability to identify complex pattern changes, the system can detect potential fault signs earlier and assist managers in formulating more targeted inspection and maintenance strategies, thereby improving the level of intelligence in equipment operation and maintenance management. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention.

[0028] Figure 2 This is a diagram showing the diagnostic results of the fault diagnosis model. Detailed Implementation

[0029] 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.

[0030] To address the issues of low accuracy, insufficient fault feature extraction, and inaccurate global change pattern measurement in fault diagnosis models for heavy engineering machinery due to the characteristics of current sensor signal data that are nonlinear, time-varying, and highly noisy, this invention aims to propose an unsupervised clustering method based on improved fuzzy C-means derived from regression and morphological fluctuation consistency offset distance. This method can effectively improve the accuracy of anomaly detection in sensor time-series data and fault diagnosis of large engineering equipment.

[0031] refer to Figure 1 This invention provides a method for fault diagnosis of heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, comprising the following steps:

[0032] Step S1: Based on fault feature extraction derived from regression, obtain the feature matrix;

[0033] The method for extracting fault features based on source regression is as follows: extract the fault features of heavy engineering equipment using a source regression model, perform linear fitting, and construct feature matrices under different fault modes.

[0034] Step S2: The feature matrix is ​​processed by the fuzzy C-means unsupervised clustering method based on the morphological fluctuation consistency offset distance to obtain the minimum target partition and partition matrix; by using the source regression model to extract fault features and performing linear fitting, feature matrices under different fault modes are constructed. On this basis, the fuzzy C-means method based on the morphological fluctuation consistency offset distance is used to calculate the fluctuation trend consistency of fault features and capture the global trend change of fault modes.

[0035] The objective function of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is expressed as follows: In the formula This indicates the dimension of the sample data; Indicates sample Compared to the first Membership degree of a class; For the Nth sample data One sample; Indicates the first The center of each category; divide N data samples into c classes; express and The formula for calculating the morphological fluctuation consistency offset distance value is as follows.

[0036] (1)

[0037] in, Representing two sequences and The relative average deviation between; Table Showing fluctuation amplitude sequence and The relative deviation; Representative sequence and The consistency of the fluctuation trend, with a value range of [-1, 1].

[0038] After constructing the objective function, the partition matrix of the feature matrix needs to be obtained. and minimum target partition This process employs an iterative optimization algorithm, and the entire optimization process is as follows:

[0039] Step S21, Initialize the partition matrix: Let ,in ;

[0040] Step S22, calculate the class center of each class, using the following formula: ;

[0041] Step S23, calculate the partition matrix The calculation formula is as follows: ;

[0042] Step S24: Iterate through steps S21 to S23, that is, after completing one step, let k = k + 1 and return to step S22; the stopping condition for iteration is... , A threshold, given by the user, represents the allowable error range; after the iteration stops, the minimum target partition is obtained. and partition matrix .

[0043] Step S3: Construct a fault diagnosis model, train the fault diagnosis model, and identify fault categories. Methods include:

[0044] In the constructed fault diagnosis model, the input of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is the feature matrix extracted from the regression model; the extracted feature matrix is ​​the coefficient matrix calculated using the regression model under various system fault states. In the formula, This represents the coefficient vector extracted under each state;

[0045] coefficient matrix The data is input into a fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance. Through iterative optimization, the optimal parameters for this method are determined, and model training is completed. After training the diagnostic model, the autoregressive coefficients derived from the regression model are extracted from the test sample data. ;make ,Will The input is fed into the trained diagnostic model to obtain... The corresponding membership matrix is ​​used to determine the fault state type of the system to which the test sample belongs, thus realizing the classification and identification of the test sample.

[0046] This invention extracts fault features from raw sensor signals using source regression, sensitively reflecting the system's operating state into the source regression model, capturing the relationships between various feature parameters, and mitigating the impact of outliers in the dataset. Subsequently, within an unsupervised clustering framework using fuzzy C-means, it employs morphological fluctuation consistency offset distance as a metric function and optimizes the objective function to calculate the consistency of the fluctuation trends of fault features, capturing the global trend changes of fault patterns. This method not only overcomes the shortcomings of existing technologies, such as unstable diagnostic performance due to insufficient feature extraction under strong noise conditions, but also accurately measures the inherent global change patterns of complex time series without requiring a large number of labeled fault samples, thereby improving the accuracy, robustness, and generalization ability of fault diagnosis and anomaly detection.

[0047] This embodiment provides a fault diagnosis method for heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, including the following steps:

[0048] (1) Data preparation

[0049] Project Background: The Nanjing Heyan Road River Crossing Tunnel Project is 5.735 km long, of which the shield tunneling section is 2885 m long. The shield tunneling machine used in this project is a slurry balance shield tunneling machine (S-1189) from Herrenknecht AG of Germany, with a cutterhead excavation diameter of 15.07 m and a composite plate structure.

[0050] Data Collection: A simulation model of the tunnel boring machine's hydraulic propulsion system was established using the AMESIM simulation platform. By modifying the simulation model parameters, fault datasets were obtained for five operating states. The fault states and their corresponding fault numbers are shown in Table 1. Parameters reflecting the system's operating state, such as hydraulic cylinder displacement, hydraulic cylinder speed, speed control valve output flow, load force, and system pressure, were extracted for each state. Ten samples were extracted for each state, each containing 60 seconds of simulation data. Six of the ten samples for each state were used as the training sample set, and four were used as the test sample set, resulting in a total of 30 training samples and 20 test samples.

[0051]

[0052] Table 1

[0053] (2) Extraction of fault features of tunnel boring machine propulsion system based on source regression

[0054] The autoregressive coefficients derived from the regression model for each fault state, calculated using the training sample set, are shown in Table 2.

[0055]

[0056] Table 2

[0057] (3) Fault diagnosis results of tunnel boring machine propulsion system based on fuzzy C-means unsupervised clustering model with morphological fluctuation consistency offset distance

[0058] The classification results of the test set samples are accurate, and each fault mode can be correctly diagnosed and identified with 100% accuracy. This demonstrates that the proposed improved fuzzy C-means unsupervised clustering model fault diagnosis method is effective and feasible for the hydraulic propulsion system of tunnel boring machines. The diagnostic results of the fault diagnosis model are as follows: Figure 2 As shown.

[0059] 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 fault diagnosis of heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering, characterized in that, Includes the following steps: Step S1: Based on fault feature extraction derived from regression, obtain the feature matrix; Step S2: The feature matrix is ​​processed by the fuzzy C-means unsupervised clustering method based on the morphological fluctuation consistency offset distance to obtain the minimum target partition and the partition matrix; Step S3: Build a fault diagnosis model, train the fault diagnosis model and identify fault categories.

2. The method for fault diagnosis of heavy engineering equipment integrating temporal pattern recognition and unsupervised clustering according to claim 1, characterized in that, In step S1, the method for extracting fault features based on source regression is as follows: extract the fault features of heavy engineering equipment using a source regression model, perform linear fitting, and construct feature matrices under different fault modes.

3. The method for fault diagnosis of heavy engineering equipment integrating temporal pattern recognition and unsupervised clustering according to claim 2, characterized in that, In step S2, the objective function of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is expressed as follows: In the formula This indicates the dimension of the sample data; Indicates sample Compared to the first Membership degree of a class; For the Nth sample data One sample; Indicates the first The center of each category; divide N data samples into c classes; express and The morphological fluctuation consistency offset distance value.

4. The method for fault diagnosis of heavy engineering equipment integrating temporal pattern recognition and unsupervised clustering according to claim 3, characterized in that, In step S2, the process of obtaining the minimum target partition and the partition matrix adopts an iterative optimization algorithm. The entire optimization process is as follows: Step S21, Initialize the partition matrix: Let ,in ; Step S22, calculate the class center of each class, using the following formula: ; Step S23, calculate the partition matrix The calculation formula is as follows: ; Step S24: Iterate through steps S21 to S23, that is, after completing one step, let k = k + 1 and return to step S22; the stopping condition for iteration is... , A threshold, given by the user, represents the allowable error range; after the iteration stops, the smallest target partition is obtained. and partition matrix .

5. The method for fault diagnosis of heavy engineering equipment that integrates temporal pattern recognition and unsupervised clustering according to claim 4, characterized in that, In step S3, the methods for training the fault diagnosis model and identifying fault categories include: In the constructed fault diagnosis model, the input of the fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance is the feature matrix extracted from the regression model; the extracted feature matrix is ​​the coefficient matrix calculated using the regression model under various system fault states. In the formula, This represents the coefficient vector extracted under each state; coefficient matrix The data is input into a fuzzy C-means unsupervised clustering method based on morphological fluctuation consistency offset distance. Through iterative optimization, the optimal parameters for this method are determined, and model training is completed. After training the diagnostic model, the autoregressive coefficients derived from the regression model are extracted from the test sample data. ;make ,Will The input is fed into the trained diagnostic model to obtain... The corresponding membership matrix is ​​used to determine the fault state type of the system to which the test sample belongs, thus realizing the classification and identification of the test sample.

Images