LAUSR

Polypropylene (PP) is one of the raw materials to produce automotive interior parts. Under the high-temperature condition, the volatile gas of PP will have a pungent odor. Therefore, an effective gas detection method should be proposed. In this work, a lightweight interleaved residual dense network (LIRD) is proposed and coupled with an electronic nose (e-nose) to classify the volatile gas of industrial PP used in the automotive interior. The LIRD combines the lightweight interleaved group convolution (LIGC) and residual dense network (RDN). The LIGC module is proposed to effectively reduce the number of parameters in the convolution calculation and realize the information exchange among channels. To avoid the feature degradation, RDN is introduced to fuse the shallow and deep features, and the parameters are trained adaptively to enhance the classification stability. By means of the self-developed e-nose system, the volatile gas of two PP materials (2240 S and 1120 K) is detected under different temperature gradients. Compared with other deep learning methods, the LIRD has a better classification result of 99.20% and 99.00%, the best F1-score is 0.9920 and 0.9900, and the best Kappa coefficient is 0.9900 and 0.9875 in two PP materials. The results show that the combination of LIRD and e-nose is demonstrated as an effective analytical technique for quality monitoring in the industrial production process of automotive interior parts.