A multimodal data sensing and feature learning-based self-adaptive hybrid approach for machining quality prediction

With the rapid development of advanced manufacturing and information technologies, smart manufacturing has been proposed and gained wide attention. As the crucial section of achieving smart manufacturing, machining quality prediction aims to use data and statistical methods to predict the quality level of products or services during manufacturing, service provision, or product delivery. However, most data-driven methods suffer from using unimodal data, ignoring the complementary of other modal data, and using a single model in one prediction task, ignoring model unstable fluctuations corresponding to different features. To bridge these gaps, a novel self-adaptive hybrid approach is proposed for obtaining high accuracy of machining quality prediction in the industrial IoT environment. Firstly, a prototype system is designed to collect and perceive multimodal data in the manufacturing workshop/system, which enables the collection and perception of more comprehensive multimodal data of physical space. Secondly, the integration of sensors, data acquisition instruments, and ontology models is used to perform the process environment, which enables the construction of information space to store and transmit multimodal data. Meanwhile, the principal component analysis and model training sensed data are used to perform the machine learning model, which enables the construction of different information models. Thirdly, a probability model matrix-based self-adaptive hybrid prediction strategy is used to perform the machining quality prediction, which enables a better fitting degree for the quality prediction task. Finally, a real-life case study and extensive experiments are conducted to verify the effectiveness and superiority of the proposed self-adaptive hybrid approach.