Sequential DTC Vector Embedding Using Deep Neural Networks for Industry 4.0

Abstract

DTC (Diagnostic Trouble Code) is the fundamental diagnostic metric through which a system will gauge the function of a module (mechanical component) of a machine. Every machine is made up of several hundreds of modules, in this paper our research tries to understand how a module is related with other modules in a machine through which we tried to understand how effectively a system can understand the relationships between DTC (diagnostic trouble codes) in terms of mechanical components of a machine using Deep Neural Networks especially sequential models such as RNN and non-sequential models such as CNN. This analysis also helps to understand the mechanics of DTC generation on dynamic conditions of a vehicle. To overcome the problems, such as DTC vector sparsity, normalize the geo-graphical conditions, irrational driving etc., we are proposing a novel mechanism called Sequential based DTC Embedding simply called SDVE (Sequential DTC Vector Embedding). SDVE is the novel technique which comprises of different techniques called component relation-based vector, which is vectorized among the various DTC (Diagnostic Trouble Code) through sequential and frequency categorization technique. We see multiple applications of this DTC relational vector such as, vehicle diagnostic system accuracy improvement, identification of relations between parts, understanding the mechanics between parts etc., To prove the proposed algorithm empirically we have built a sample deep learning model which embedded the SVDE as a layer in a part failure prediction deep neural network architecture. Empirically results of the deeply learned model using Gated recurrent unit network for module diagnosis system shown that the SDVE (separate latent space of DTC built using Convolutions) use as an embedded layer, boosted the accuracy in part failure prediction. Empirically proven results are shown in below, which supports our research in terms of novelty and accuracy.