Mechanical equipment fault diagnosis method based on deep learning

Abstract

The invention discloses a mechanical equipment fault diagnosis method based on deep learning. The method specifically comprises the following steps of S1, carrying out the data collection and preprocessing of a main data source and a secondary data source of mechanical equipment, and obtaining a data set; S2, a five-fold cross validation method being adopted to divide the data set into a trainingset, a validation set and a test set; and S3, establishing a fault diagnosis model based on the CNN and the BD-LSTM, inputting the training set into the fault diagnosis model, extracting hidden features, performing training, and outputting a diagnosis result. According to the method, BD-LSTM is adopted to perform smooth tracking and result prediction, and uncertainty caused by operation and environmental interference is processed, sensor monitoring data adopts CNN and BD-LSTM to extract hidden features in parallel, output of two irrelevant paths can influence prediction, and each parameter inthe network can be corrected according to predicted errors.

Description

technical field [0001] The invention relates to the technical field of system fault diagnosis, in particular to a method for fault diagnosis of mechanical equipment based on deep learning. Background technique [0002] In intelligent manufacturing, the intelligent operation and maintenance and health management of equipment will inevitably penetrate directly into the operation management of the enterprise and even the entire life cycle of the product, thereby reducing the loss of the enterprise and affecting the decision-making of the enterprise. One of the key elements of the new model of intelligent manufacturing is "remote operation and maintenance service". (Products) Remote unmanned control, early warning of working environment, monitoring of operating status, fault diagnosis and self-repair, etc. [0003] Fault diagnosis technology plays a vital role in large-scale systems, especially industrial systems. It can obtain the fault model of the diagnostic object through d...

AI Technical Summary

Problems solved by technology

[0004] (1) Fault diagnosis methods based on physical models largely depend on the quality of the model, and equipment components are closely related to uncertain working conditions, and their degradation process is usually nonlinear, so the application of this type of method is very limited;

[0005] (2) Experience-based fault diagnosis methods mainly use empirical feedback data (such as failure time) to adjust the parameters of some analysis models (such as Weibull distribution, exponential distribution), and estimate the failure time of the system through these parameter models, but there are often predictions Insufficient precision, excessive reliance on knowledge and experience, etc.

Although the two patent documents with the publication number CN108645615A and the publication number CN109343505A have considered the problem of time dependence, the mean square amplitude is used in the feature extraction of the fatigue state of mechanical equipment, and the accuracy of this method is not enough , will lead to problems in the extracted features and affect subsequent predictions, and the above technical solution also has the following problems:

[0007] (1) Although recurrent neural network (RNN), long-short-term memory network (LSTM) or memory units are added to each node, the feature extraction part is still using the traditional method, which will not only affect the accuracy of feature extraction, but also affect the accuracy of feature extraction. Correctness of subsequent health state predictions

[0008] (2) The above scheme is only aimed at a single sensor signal, and the health status of the actual mechanical equipment is not only related to real-time monitoring data, but also related to its operating parameter data and inspection data. The information that needs to be considered in the above scheme is not comprehensive enough

[0009] Based on the above defects, it can be seen that the feature extraction part of the existing technology is still using the traditional method, which is not accurate enough, which will not only affect the accuracy of feature extraction, but also affect the correctness of subsequent health status prediction; and the collected data relies on sensor signals, there is no Taking into account information such as operating parameter data and inspection data

Images