| Driven by the fourth industrial revolution,“Made in China 2025” and other relevant policies,permanent magnet synchronous motor(PMSM)has been widely used in new energy wind power generation,new energy vehicles,advanced rail transit equipment,aerospace,robotics,and other major national development fields due to their advantages in structure and performance.A safe and reliable high-performance PMSM system is the core and key equipment in the aforementioned fields,and the demand is urgent.Although the fault diagnosis technology for PMSM has achieved certain results.However,with the expansion of PMSM application fields and the continuous development of related equipment towards large-scale,highly integrated,and diversified functions,new challenges have been brought to the fault diagnosis field of PMSM.Therefore,the development of advanced intelligent fault diagnosis and monitoring technology for PMSM has important theoretical and practical value for improving the overall performance and service life of the high-precision and cuttingedge equipment,and promoting the development of major strategic equipment in China.This paper aims at the key technical problems such as the difficulty in obtaining fault data in industrial applications,the complexity of feature extraction and hyperparameter adjustment in variable working conditions,and the shortage of training samples under small samples,which are faced by PMSM fault diagnosis.Focused on the research of PMSM intelligent fault diagnosis technology based on data image features.From the perspective of data images,artificial intelligence algorithm is used to automatically extract,learn,and apply knowledge to view the equipment fault status like human.The innovation and main content of this article are as follows:(1)This paper researched and developed a simulation experimental platform for PMSM faults.Firstly,addressing the problem of parameter mismatch and inconsistency in simulating a megawatt wind power generation system using a small power motor,the wind turbine model of the megawatt wind power generation system and a virtual wind power generation system model were established.Simulation strategies for static and dynamic operating characteristics were formulated.Secondly,based on the analysis of PMSM operating characteristics in the drive system,control strategies for PMSM output control and flexible adjustable load simulation in the drive system were devised.Furthermore,through fault analysis and motor structure analysis,implantation schemes for various PMSM faults were designed,and corresponding fault prototypes were developed.Finally,based on the wind power generation system simulation and drive system simulation strategies,a PMSM fault simulation experimental platform was developed,and the effectiveness of the proposed operating characteristic simulation techniques was verified through experiments.This provides experimental and testing conditions for studying fault diagnosis of PMSMs in megawatt wind power generation systems and electric drive systems.(2)A demagnetization fault diagnosis and state recognition method based on local features of image is proposed.Firstly,the symmetrized dot pattern(SDP)method is used to convert the one-dimensional magnetic flux leakage signal into two-dimensional SDP image,which mines the fault features of the two-dimensional image in the magnetic flux leakage signal.Secondly,a deep feature extraction and encoding method for images is proposed,which unifies the dimensions of shallow local features and reduces the computational cost of fault diagnosis.At the same time,the adaptability and anti-disturbance performance of fault diagnosis is improved.Finally,an improved autonomous learning multiple-model classifier is proposed to avoid the complex hyperparameter adjustment of traditional intelligent classifiers under different working conditions.The proposed method solves the difficulty of extracting one-dimensional signal features and the complex hyperparameter tuning problem of diagnostic model in traditional methods.(3)A fault diagnosis method for permanent magnet synchronous motors based on image retrieval is proposed.Inspired by visual perception and fingerprint recognition,this method first utilizes two-dimensional SDP images to reveal the characteristics of local demagnetization faults and interturn short-circuit faults.Secondly,the KAZE-BF encoding technique is proposed to reduce the computational burden of complex feature vectors in image content retrieval,which greatly simplifies the feature vectors of images.Finally,a cyclic update algorithm based on image fusion technology is designed to utilize the spatiotemporal correlation and information complementarity of images under different operating conditions.Thus,the fault diagnosis performance of the proposed method under different operating conditions is enhanced.The proposed method solves the problems of insufficient adaptability of traditional PMSM fault diagnosis methods to operating conditions and the large number of samples required for model training.(4)A new intelligent fault diagnosis framework for PMSM under small industrial sample conditions is proposed.Firstly,a custom phase space reconstruction(CPSR)method is proposed to reconstruct the magnetic flux leakage signal into a twodimensional image,and CPSR image is used to represent the health status of PMSM,which reduces the complexity of fault diagnosis and reveals the two-dimensional global fault characteristics hidden in the magnetic flux leakage signal.Secondly,a parallel semi-supervised stacked autoencoder diagnostic architecture with weighted decision fusion strategy is constructed,which makes full use of the information of unlabeled samples through semi-supervised learning and improves the robustness of diagnosis.Finally,the effectiveness of the proposed method is verified through a large number of rich experiments.This article closely revolves around the common demand for high-quality service quality of high-precision and cutting-edge equipment in major development fields in China.Taking PMSM,which is widely used in modern power generation and transmission systems,as the object,this article focuses on researching the simulation of applied equipment characteristics,image representation of signals,fault feature extraction,and fault diagnosis methods.Perceiving the overall and local features of device status data from the perspective of Data image to obtain knowledge of device operation status is proposed.One-dimensional data is expanded to two-dimensional space,which contributes to achieving PMSM fault diagnosis through the expression,reasoning,learning,and application of data image.A highly intelligent condition monitoring and theoretical and methodological system for the fault diagnosis of entire process is established,including PMSM operation state perception,image expression,information extraction,and automatic diagnosis.The urgent bottleneck issues faced by PMSM fault diagnosis,such as the lack of experimental testing environment,improvement of stability and accuracy,adaptability to multiple operating conditions,and small sample conditions are solved.This paper provides the relevant theoretical and technical guidance for promoting the fault diagnosis and maintenance of PMSM in new energy wind power generation and electric drive. |
