Research On Steering Gear Detection Technology Based On Deep Learning

As a servo drive,the steering gear is widely used in aerospace,ships and other fields,and plays an important role in missile attitude transformation and heading control.In order to ensure the stability of the performance,the research on the accurate and automatic detection technology of the abnormal state of the steering gear is of practical significance.At present,the automation of steering gear testing equipment and the simplicity and reliability of parameter testing are available,but manual analysis and decision-making of massive data is time-consuming and labor-intensive,and accuracy is difficult to guarantee.This paper introduces deep learning technology to build a steering gear anomaly detection model,realizes automatic feature extraction and anomaly identification of steering gear performance parameter data,and promotes the development of steering gear detection technology in the direction of automation and intelligence.First,the paper constructs the GWO-DNN-LRC model based on Deep Neural Networks.Specifically,in order to improve the classification performance of the model,the powerful feature extraction ability of the Deep Neural Networks is combined with the classification performance of the Logistic Regression Classifier,and the Grey Wolf Optimization algorithm is used to fine-tune the number of nodes in the hidden layer of the Deep Neural Network.The experiments involve the influence of the number of hidden layers on the classification performance of the Deep Neural Networks,the optimization of the hyper-parameters of the Deep Neural Networks,and the selection of the steering gear anomaly detection model classifier.In the performance comparison with LRC,DNN and GWO-DNN models,the proposed model GWO-DNN-LRC demonstrated high quality anomaly detection ability,with an average 99.261% accuracy,98.417% precision,98.062% recall and 98.217% F-Score for 5repeated experiments.It mainly solves the problem that the Logistic Regression Classifier is significantly affected by the data sample size,and the problem that the Deep Neural Networks hyper-parameters are difficult to select.Second,the HPSOGWO-CNN model is constructed based on the Convolutional Neural Networks.Specifically,the hyper-parameters of Convolutional Neural Networks are finetuned by using a hybrid algorithm,namely HPSOGWO algorithm,to improve the classification performance.Compared with Deep Neural Networks,the addition of convolution layers,pooling layers,and dropout layers in Convolutional Neural Networks make the performance of the network more powerful.The experiments mainly analyze the feasibility of applying the HPSOGWO algorithm to the fine-tuning of neural network hyperparameters.Through the performance comparison analysis of ten models（KNN,SVM,BP,CNN,PSO-CNN,GWO-CNN,MGWO-CNN,Wd GWO-CNN,RW-GWO-CNN,HPSOGWO-CNN）,it shows that the proposed model has the best level and excellent performance.HPSOGWO-CNN model can achieve 99.846% accuracy,99.748% precision,99.498% recall,99.618% F-Score and 0.99565 Kappa.In the classification performance evaluation of different categories,Seven of the eleven categories achieved 100% accuracy,precision and F-score.The model mainly improves the classification of small samples,and it is concluded that the HPSOGWO algorithm is an excellent automatic hyper-parameters selection technology.This article builds two steering gear anomaly detection models from different perspectives.The construction of the model effectively solves the problem that small samples in the steering gear test data are difficult to be accurately classified,and is suitable for deep feature extraction and anomaly detection of sample imbalanced data.The research of steering gear detection technology based on deep learning reduces the workload of manual data analysis and decision-making,realizes the automation of steering gear anomaly detection,and provides guarantees for the security of aircraft and ships during navigation.