Edge-computing based soft sensors with local Finite Impulse Response models for vehicle wheel center loads estimation under multiple working conditions

Wheel center loads are usually required to be measured for the durability assessment of vehicle key components, and thus play an important role in the reliability design of vehicles. The current method of obtaining these loads is to install sophisticated and expensive sensors on prototype vehicles to conduct direct measurement. However, it is economically infeasible to equip many prototype vehicles simultaneously for road testing. The paper aims to design and implement edge-computing based soft sensors to estimate these wheel center loads in real time. A set of Finite Impulse Response (FIR) models are trained offline under multiple working conditions, and a Maximum Likelihood Estimation (MLE) classifier is used to determine which working condition the vehicle currently is under so that the wheel center loads can be estimated accurately. The soft sensors are implemented with an edge computing paradigm, and the whole process of data collection, pre-processing, calculation, and upload runs automatically on edge devices. Experimental results show that the estimate from the proposed method fits the measurement well, and the edge-computing based soft sensors meet the real-time requirements for wheel center load estimation. •A data-driven soft sensor is designed and developed to estimate wheel center loads.•An edge computing paradigm is proposed to deploy and manage soft sensors efficiently.•Real road test shows the high performance and efficiency of the soft sensors.