Neural Network Approach to Estimate the Performance of Processes Involved in Vehicle Service 2015-26-0043
Regular service of the vehicle is to be done with high precision service equipment, to ensure the factory performance of the vehicle over the entire life of product usage. However, complex nature of the physical processes involved in the service of the vehicle subsystems makes it costly for optimizing the service equipment performance for entire range of operation. Air-conditioning service (ACS) equipment is one such product in the diagnostics domain which deals with compressible, transient and two phase flow in open loop systems.
Development of control system for the service equipment to perform optimally over the entire operational range requires accurate mathematical model of the system under study. Application of mathematical model based approach requires calculation of geometrical details, environment information and fluid properties during the process for estimating the process behavior. Generating these empirical details may require intensive testing with usage of advanced instrumentation which makes this approach more complex, time consuming and costly. And this may not be feasible in the available time to market the design solution.
Neural Network (NN) can be applied for modeling complex nonlinear systems that are not easily modeled with a closed-form equation. This approach requires correct input-output model of the process which specifies the right input and output parameters along with the environmental parameters over the entire range of operation. It doesn't require accurate intermediate details of the process.
This paper presents a practitioners approach for applying a multilayer Neural Network based model with back propagation algorithm for optimization of processes involved in vehicle service. Refrigerant recharge process in ACS equipment is taken as a case study to explain the approach. Statistical technique (Design of Experiments) is used for minimizing the test data points which are required for training the NN. As a result, optimized weightage of the affecting parameters is obtained. This NN model can be used to estimate the performance of the physical process by entering the required operating set as input values.