Research On The Identification Of Arc Furnace Smelting State Characteristics Based On Arc Sound Signa

Electric arc furnace is an important equipment in the field of iron and steel smelting and its production is increasing year by year,to achieve the electric arc furnace smelting steel quality,increase capacity and reduce energy consumption,reduce pollution is the development of steel smelting industry objective needs and urgent requirements.The timely and accurate identification of the smelting state of electric arc furnace can optimize the smelting process to improve smelting efficiency and reduce energy consumption per ton.Arc sound,as an accompanying signal of electric arc combustion,can be obtained non-contact and has a strong ability to map the smelting state of electric arc furnace.Therefore,it is important to identify the smelting status of electric arc furnace based on arc sound.In this paper,we analyze the characteristics of the de-noised sound signal from the time domain and frequency domain,and use the sound signal to identify the smelting state of electric arc furnace.The main research contents are as follows:1.Choose to take the 75-ton steel furnace cold material smelting experimental platform as an example for research.Build the sound signal acquisition hardware platform and design the human-computer interaction software.According to the German metallurgical industry sound research protocol to select signal acquisition points,under strong electromagnetic and strong noise interference,collect the electric arc furnace smelting sound signal and pre-process the collected sound.Combining the worker introduction,smelting state classification theory,arc current curve to classify smelting state and label the smelting state,the sound database based on the electric arc furnace arc sound is established.2.To propose a denoising algorithm and a practical denoising method to improve the denoising effect for the requirements of electric arc furnace arc sound denoising.Analyze the factors affecting the denoising effect,and study the traditional wavelet threshold denoising and its improvement function.Taking the arc sound of electric arc furnace smelting state identification as the research background,an adjustable parameter improvement wavelet threshold denoising algorithm is proposed for the sound change characteristics in the electric arc furnace smelting process.Based on the idea of weighted averaging,the parameters α and μ,are introduced to adjust the approximation speed to solve the reconstruction oscillation problem.Through theoretical analysis,its asymptotic property and the characteristics when special values are taken are proved.In addition,the root-mean-square error is used to prove the left and right channel errors,and for the characteristics of more interference noise,the method of denoising and averaging the two channels separately is proposed to target the noise reduction process of the arc sound signal and improve the denoising robustness.Finally,the root-mean-square error and signal-to-noise ratio indicators are used to evaluate the denoising effect,and the speech spectrum graph is drawn to visually display the denoising effect.Through the improvement,the noise removal effect is obvious.3.Extract a variety of features and analyze them,and vector quantize some of them.From the perspective of time domain,frequency domain,inverse spectral domain and speech spectrogram,short-time average amplitude,short-time autocorrelation function,spectral center of mass,power spectral density and other feature indicators are selected to characterize the arc sound of different smelting states.The differences in the characteristics of different smelting states are compared from the perspectives of average value,peak value,peak-to-valley ratio,etc.to reveal the inner connection between the sound signal changes and the five smelting states of arc starting,well penetration,electrode lifting,composition adjustment,and stable temperature rise.Extract the inverse spectrum features MFCC,GFCC,and vector quantization of MFCC using LBG algorithm.The three-dimensional speech spectrum diagram is drawn to express the sound states in three dimensions,and the two-dimensional speech spectrum diagram is drawn using two algorithms to visually describe the sound differences between different smelting states.4.Build a model for state identification,and combine the identification results with feature analysis for parameter optimization,GFCC,vector quantization MFCC,and deep learning feature multi-feature fusion algorithms are proposed for improvement..Support vector machine(SVM)model and convolutional neural network(CNN)model are built,MFCC,vector quantized MFCC,and GFCC are selected for smelting state recognition simulation experiments using support vector machine,and smelting state recognition simulation experiments using convolutional neural network model for speech spectrogram input.Based on the feature analysis and single feature simulation experiments smelting state between orientation difference analysis,derive deep learning features,using inverse spectrum features and deep learning features for fusion algorithm state recognition.Using the decision fusion and feature fusion algorithms,the PCA dimensionality reduction algorithm is introduced to optimize the smelting state recognition algorithm for the problem of too long feature dimension in the fusion algorithm.The model is trained with 2000 training sets of data and validated with another 2000 test sets of data hidden label simulation,and the highest recognition rate of five smelting states combined reaches 96.6%.