Fault Diagnosis Based On Denoising Orthogonal Auto-encoder In TE Process

With the development of industrial technology,the scale of control system is expanding,and the internal structure is also becoming more and more complex.Any abnormal operation process may bring unpredictable safety and property losses.With the coming of the era of industrial 4.0,each section of the production system turns to be data-oriented and highly intelligent.How to make effective use of the massive data related to the system and make quick and accurate detection of faults is the focus of this paper.At present,the traditional statistical methods of process monitoring are faced with many limitations: the data in the actual process doesn't subject to the ideal distribution,unstable condition of production equipment and other complex factors,such as physical environment variables,sensor measurement errors,etc.Based on these research background and requirements mentioned above,this paper investigates various fault detection and diagnosis methods,and focuses on the algorithm based on auto-encoder algorithms.According to the actual situation,existing methods are improved with orthogonal limits and denoising tricks,thus an improved auto-encoder algorithm is proposed and verified in the simulation experiments of chemical Tennessee Eastman Process.The details of work in this article are as follows:1)Fault detection algorithm based on Denoising Orthogonal AutoEncoderThe fault diagnosis of large systems is a very complex task,and the learning ability of the traditional fault detection algorithm is extremely limited.As the data itself has certain redundancy and noise,the orthogonality as well as the noise reduction are introduced into auto-encoder algorithm,which is the proposed Denoising Orthogonal Auto-Encoder(DQAE)to overcome the shortcomings of auto-encoder algorithm and enhance the ability of learning and extraction key features.Through the application in Tennessee Eastman process simulation experiment,the proposed method has lower false positive rate and missing detection rate compared with the traditional fault detection method,which verifies the effectiveness and superiority of this method.2)Fault diagnosis algorithm based on Multi-fault Denoising Orthogonal Auto-EncoderWith the complexity and large-scale evolution of the system,not only the consequences of faults are serious,but also the difficulty of maintenance and the cost are very large.The real-time online fault detection is becoming more and more important to the system,and the requirements for the accuracy and speed of fault diagnosis are also getting higher and higher.In this paper,DQAE algorithm is applied as a sub-model for each type of fault feature extraction and then aggregated into Multi-fault Denoising Orthogonal Auto-Encoder(MDQAE)method.Through the normalization of the statistics and threshold of the model output,the test data statistics are compared with each threshold to determine class of the fault,which can improve the accuracy and reliability of fault diagnosis.Experimental results show that this scheme has lower false positive rate and missed detection rate for most faults,and reduces the complexity of its implementation.It provides possibility for online fault diagnosis,and is also easy to be applied and popularized in engineering.