Acoustic emission monitoring of sawing process: artificial intelligence approach for optimal sensory feature selection

Abstract

A methodology is presented for acoustic emission (AE) monitoring of Douglas fir wood in circular sawing process under extreme cutting conditions. An AE sensor was mounted on the saw guide to investigate the blade vibration, the interaction between the saw and workpiece, and the cut surface waviness. The acquired signal was filtered using a wavelet de-noising method. Signal processing was performed in both the time and frequency domains, and different features were extracted. The effects and significance of cutting parameters such as feed speed, rotation speed, and depth of cut on the AE signal were discussed. Both the statistical regression and artificial intelligence approach were employed to correlate the AE-extracted features with the cutting power and waviness. Adaptive neuro-fuzzy inference system (ANFIS) was used to predict the cutting power and waviness. Optimal feature selection was performed by combining the ANFIS with a metaheuristic optimization algorithm. Particle swarm optimization (PSO) method was used to find the AE-selected features that result in a minimum ANFIS model error. ANFIS was trained using both the PSO and a hybrid algorithm. The results indicate the effectiveness of the wavelet de-noising method for signals that are polluted by different low- and high-frequency sources. AE signals were highly influenced by the saw blade deflection and its vibration response. ANFIS modeling showed higher performance than statistical regression for cutting power and waviness prediction. Optimal feature selection improved the model accuracy, which was further enhanced when the optimal sensory features were combined with the studied cutting parameters. The optimal ANFIS performance was achieved when the network was trained using PSO algorithm. The presented methodology can successfully monitor the cut surface waviness.