Online Self-adapting Detection On The Dynamic Characteristics Of Arc Welding Power Supply Based On The Coupling Dimension Reduction Of Noise, Correlation And The Time Consumed

The application of various electric arc welding has boosted the industrial development; atthe same time, the quality of the characteristics of arc welding power supplies has become oneof the key factors affecting the quality of industrial products. Therein, dynamic characteristicsof arc welding power supplies are especially important, which affecting the quality of welding,efficiency and even cost. Large numbers of scholars domestic and overseas have launchedrelevant researches, however, it has not yet been reported that any detection technology ofdynamic characteristics of arc welding power supply has ever been put into practice. On onehand, the welding process is complicated, and it's hard to achieve practical automaticdetection technology; more importantly, with the development of arc welding technology, newarc welding power supplies of varied purposes emerge endlessly. Applying the usual methodsdemands a deep study on a specific type and establishing a corresponding detection system,which are undoubtedly costly and impractical.Therefore, this subject introduces the theory of dimension reduction to crack problemsderived from detecting diversified arc welding power supplies. The basis of this theory is tocollect sufficient features related to the dynamic characteristics of arc welding power supplies,and to build a relatively large feature library which allows to extending. When applied to aspecific type or specific model of arc welding power supply, judging by experiments, thismethod is able to automatically seek out feature subsets which can effectively calculate andevaluate the quality of the power supply dynamic characteristics from the feature library.During the calculation process, noise, correlation, and the time consumed are simultaneouslytaken into consideration.A relatively complete experimental platform has been established, whose core function isto accurately collect the output voltage and current signals during the welding process of arcwelding power supplies. This experiment demands the participation of a series of instruments,including a self-developed wavelet analyzer.On the basis of summarizing research results brought by predecessors, this thesisinnovatively put forward a series of features, such as "repetition of normal circles", "coefficient variation of intervals between voltage autocorrelation function peaks", whichwere also testified through experiments. A feature library involving65features wasestablished, which covers CO2welding and P-GMAW. This is an open library, so the weldingtypes it involves will extend with the development of researches and application, and thenumber of features in the library will also increase.Questions about correlation and the time consumed during the arc welding process werealso taken into consideration, therefore, a method of coupling dimension reduction ofcorrelation and the time consumed was put forward, which can bring self-adapting onlinedetection on the dynamic characteristics of diversified arc welding power supplies. Detailedexperimental analyses have been individually done on CO2welding, P-GMAW, andDP-GMAW. Results of the experiments showed that this method not only could effectivelydetect arc welding types that were, but also those not included.Problems of noise interference in industrial manufacturing environments were also takeninto consideration, and a method of coupling dimension reduction based on noise, correlationand the time consumed was put forward to achieve the self-adapting detection of diversifiedarc welding power supplies. This method is also called the method of dimension reductionapplied to industrial detection targets, which has comprehensively taken noise interference,the time consumed during online detection, and the correlation between features intoconsideration. Results of experiments done on CO2welding, P-GMAW, DP-GMAW showedthat this method, being able to effectively detect both arc welding types included and thosenot included, could effectively solve the detection problems of diversified arc welding powersupplies.At last, the way of thinking was changed. Instead of applying the method of featureselection, feature transformation was adopted. On the basis of training and testing sets, anumber of new features, closely related to the grades of the evaluation targets, were generatedfrom the above-mentioned feature library. Results of experiments showed that the methodbased on FLDA, achieving higher classification accuracy than direct classification or themethod based on PCA did, was more preferred to generate new features to realize self-adapting detection. Compared to the above-mentioned coupling dimension reduction, thedefect of this method is that it brings lower classification accuracy, but it excels at its highspeed. When the model is established for temporary use, the method based on FLDA is morepreferred.