Data-driven technological life prediction of mechanical and electrical products based on Multidimensional Deep Neural Network: Functional perspective

With the acceleration of technology upgrading, the phenomenon of scrapping mechanical and electrical (ME) products due to the end of technological life is increasingly severe, resulting in a large amount of resource and energy waste. The change of technological life is caused by the invisible degradation of product function performance. The challenges lie in identifying the drivers of affecting technological life and building the prediction model. This paper proposes a data-driven technological life prediction method based on deep learning. Innovatively based on the substance-field theory, the leading role of "field" on the technological life of ME products in service is revealed, and the drivers that directly affect the technological life by describing technical performance as product functional characteristics are obtained. Further, using the Axiomatic Design Theory, the drivers of technological life indirectly affected are identified by characterizing external demands as functional parameters of product features. Consequently, the driver system of technological life from two levels of product itself and external demands is established. Considering the driver differences of technological life of various ME products working at the complex conditions, as well as the characteristics of large amount of driver data and strong nonlinear time variability, the Multidimensional Deep Neural Network (MDNN) model is built. MDNN is integrated by two parallel paths (one Temporal Convolutional Network and one Bidirectional Long Short-term Memory network) to fully mine the spatial and temporal characteristics in the technological life data. In addition, to deeply improve the generalization ability of the model, the operating condition data is also used as the input sequence of MDNN model to realize the technological life prediction under variable conditions. Taking a smart phone as an example, the effectiveness of the proposed method is verified. For sustainability of resources, this method can serve for the design and development direction of high-tech products, and their active remanufacturing for life cycle management.