CN103743477B - Method and device for detecting and diagnosing mechanical faults - Google Patents

Abstract

The invention discloses a kind of mechanical fault detection diagnostic method, its concrete steps are: in detected plant equipment, install multiple mechanical vibration sensor; Collection machinery equipment runs well respectively, and the vibration signal operated under different faults type condition; Generate the svm classifier rule of mechanical fault detection; SVM training aids is trained, the training rules of adjustment SVM training aids; By vibration transducer collection machinery vibration signal, and carry out fault Preliminary detection by SVM classifier; Application multisample Voting Algorithm, carries out ballot to the result of Preliminary detection and analyzes, obtain final failure detection result.A kind of mechanical fault detection equipment, is characterized in that it comprises SVM training aids and multiple mechanical oscillation signal pick-up transducers.A kind of mechanical fault detection diagnostic method proposed by the invention and equipment thereof, ensure that the accuracy of testing result, can to typical mechanical fault detection and identification, and the fault detection and diagnosis of the plant equipment to multiple variety classes, different performance feature can be adapted to, improve the precision of mechanical fault detection.

Description

Method and device for detecting and diagnosing mechanical faults

technical field

The invention relates to a mechanical fault detection method and detection equipment thereof.

Background technique

At present, there are many mechanical fault diagnosis methods, including static-based mechanical fault diagnosis methods and dynamic mechanical fault diagnosis methods. Among them, the mechanical fault detection method that takes mechanical vibration as the detection object is currently a new method in the field of mechanical fault detection. Mainstream method. However, the accuracy of fault diagnosis and the type of fault identification based on mechanical vibration fault detection methods largely depend on the analysis and processing capabilities of mechanical vibration signals. At present, most of the existing mechanical fault diagnosis methods focus on the acquisition and analysis algorithms of mechanical vibration signals, which also include a large number of different transformations or signal conversions for the collected mechanical vibration signals. And so on, in order to improve the identification ability of mechanical vibration signal. However, the failure time and diagnosis algorithm designed by this series of typical mechanical fault diagnosis methods depend largely on statistics or probability analysis, and there is no clear basis for fault identification. Because of this, the current mechanical fault detection methods have strong limitations, that is, for different scenarios and different detection objects, their fault detection capabilities and effects often vary greatly.

Contents of the invention

The technical problem to be solved by the present invention is to provide a mechanical fault detection and diagnosis method and its equipment, which can detect and identify typical mechanical faults, and can adapt to the fault detection and diagnosis of various types of mechanical equipment with different performance characteristics .

In order to solve the above-mentioned technical problems, the present invention provides a method for detecting and diagnosing mechanical faults, the specific steps of which are:

The first step: install multiple mechanical vibration sensors on the mechanical equipment to be tested, and collect various types of mechanical vibration signals for multiple different vibration points;

The second step: Let the tested mechanical equipment start to run, respectively collect the vibration signals of the normal operation of the mechanical equipment and the operation under different fault types, and obtain the original vibration data of the tested mechanical equipment;

The third step: use the SVM trainer to train it to train the sampled mechanical vibration signal, and generate the SVM classification rules for mechanical fault detection;

The fourth step: use the mechanical failure detection results as the input data of the SVM trainer, train the SVM trainer, and adjust the training rules of the SVM trainer;

Step 5: When detecting the detected mechanical fault, first collect the mechanical vibration signal through the vibration sensor, and perform preliminary fault detection by the SVM classifier;

Step 6: After multiple SVM classifiers detect and obtain multiple sets of preliminary fault detection results, apply the multi-sample voting algorithm to conduct voting analysis on the preliminary detection results to obtain the final fault detection results.

The detection results obtained each time will be used as the input data of the SVM trainer to realize the adjustment of classification rules for mechanical fault diagnosis, so that the mechanical fault diagnosis method proposed in the present invention has self-learning ability and can be corrected by using historical detection data. The fault detection method; the mechanical fault diagnosis method uses multiple sets of vibration signals as the detection data source, which can cover various vibration signals of the mechanical object to be detected, and comprehensively carry out sampling and analysis processing to ensure the accuracy of the detection results.

The multi-sample voting algorithm is a weight-based voting algorithm.

A mechanical fault detection device, which includes a SVM trainer and a plurality of mechanical vibration signal acquisition sensors, the mechanical vibration signal sampling sensors are arranged at a plurality of different measurement points on the mechanical equipment to be detected, so as to obtain various detection objects during mechanical operation The vibration signal; the mechanical vibration signal acquisition sensor sends each group of mechanical vibration signals to the SVM trainer respectively, and through the training and analysis of a large number of mechanical vibration signals, the classification rules of the SVM trainer are formed. When the mechanical fault diagnosis is actually performed, By deploying multiple sets of mechanical vibration measurement sensors on mechanical equipment, the vibration signals during mechanical operation are sampled separately, and the detection and identification of mechanical vibration signals are realized through the SVM classifier. After each SVM classifier detects and obtains a detection result , and then the voting selector analyzes the multiple sets of detection results obtained from the detection and votes to obtain the final detection result.

A mechanical fault detection and diagnosis method and its equipment proposed by the present invention use multiple sets of vibration signals as the data source for detection, which can cover various vibration signals of the mechanical object to be detected, and comprehensively perform sampling and analysis processing to ensure The accuracy of the test results can detect and identify typical mechanical faults, and can adapt to the fault detection and diagnosis of various types of mechanical equipment with different performance characteristics, which improves the accuracy of mechanical fault detection. At the same time, it also has automatic Learning ability, able to use historical detection data to revise fault detection methods.

Description of drawings

FIG. 1 is a schematic flow chart of a method for detecting and diagnosing a mechanical fault proposed by the present invention.

Figure 2 is a schematic diagram of the correction of the classifier by the mechanical fault detection result.

Figure 3 is a flowchart of the multi-sample voting algorithm.

detailed description

Referring to accompanying drawing, a kind of mechanical fault detection and diagnosis method, its specific steps are:

The first step: install multiple mechanical vibration sensors on the mechanical equipment to be tested, and collect various types of mechanical vibration signals for multiple different vibration points;

The second step: Let the tested mechanical equipment start to run, respectively collect the vibration signals of the normal operation of the mechanical equipment and the operation under different fault types, and obtain the original vibration data of the tested mechanical equipment;

The third step: use the SVM trainer to train it to train the sampled mechanical vibration signal, and generate the SVM classification rules for mechanical fault detection;

The fourth step: use the mechanical failure detection results as the input data of the SVM trainer, train the SVM trainer, and adjust the training rules of the SVM trainer;

Step 5: When detecting the detected mechanical fault, first collect the mechanical vibration signal through the vibration sensor, and perform preliminary fault detection by the SVM classifier;

Step 6: After multiple SVM classifiers detect and obtain multiple sets of preliminary fault detection results, apply the multi-sample voting algorithm to conduct voting analysis on the preliminary detection results to obtain the final fault detection results.

The detection results obtained each time will be used as the input data of the SVM trainer to realize the adjustment of classification rules for mechanical fault diagnosis, so that the mechanical fault diagnosis method proposed in the present invention has self-learning ability and can be corrected by using historical detection data. The fault detection method; the mechanical fault diagnosis method uses multiple sets of vibration signals as the detection data source, which can cover various vibration signals of the mechanical object to be detected, and comprehensively carry out sampling and analysis processing to ensure the accuracy of the detection results.

The multi-sample voting algorithm is a weight-based voting algorithm.

A mechanical fault detection device, which includes a SVM trainer and a plurality of mechanical vibration signal acquisition sensors, the mechanical vibration signal sampling sensors are arranged at a plurality of different measurement points on the mechanical equipment to be detected, so as to obtain various detection objects during mechanical operation The vibration signal; the mechanical vibration signal acquisition sensor sends each group of mechanical vibration signals to the SVM trainer respectively, and through the training and analysis of a large number of mechanical vibration signals, the classification rules of the SVM trainer are formed. When the mechanical fault diagnosis is actually performed, By deploying multiple sets of mechanical vibration measurement sensors on mechanical equipment, the vibration signals during mechanical operation are sampled separately, and the detection and identification of mechanical vibration signals are realized through the SVM classifier. After each SVM classifier detects and obtains a detection result , and then the voting selector analyzes the multiple sets of detection results obtained from the detection and votes to obtain the final detection result.

A mechanical fault detection and diagnosis method and its equipment proposed by the present invention use multiple sets of vibration signals as the data source for detection, which can cover various vibration signals of the mechanical object to be detected, and comprehensively perform sampling and analysis processing to ensure The accuracy of the test results can detect and identify typical mechanical faults, and can adapt to the fault detection and diagnosis of various types of mechanical equipment with different performance characteristics, which improves the accuracy of mechanical fault detection. At the same time, it also has automatic Learning ability, able to use historical detection data to revise fault detection methods.

Claims (2)

1. a mechanical fault detection diagnostic method, its concrete steps are:

The first step: install multiple mechanical vibration sensor in detected plant equipment, gathers the mechanical oscillation signal of number of different types for multiple different oscillation point;

Second step: allow detected plant equipment start running, collection machinery equipment runs well respectively, and the vibration signal operated under different faults type condition, obtains the vibration raw data of detected plant equipment;

3rd step: application SVM training aids train its to the mechanical oscillation signal obtained of sample train, the svm classifier generating mechanical fault detection is regular;

4th step: using the input data of mechanical fault detection result as SVM training aids, SVM training aids is trained, the training rules of adjustment SVM training aids;

5th step: when to detected mechanical fault detection, first by vibration transducer collection machinery vibration signal, and carry out fault Preliminary detection by SVM classifier;

6th step: after multiple SVM classifier detection obtains the preliminary failure detection result of many groups, application multisample Voting Algorithm, carries out ballot to the result of Preliminary detection and analyzes, obtain final failure detection result.

2. a kind of mechanical fault detection diagnostic method according to claim 1, is characterized in that, described multisample Voting Algorithm is a kind of Voting Algorithm based on weight size.