Performance Prediction of an Automotive Assembly Line Based on ARMA-ANN Modeling

Performance Prediction of an Automotive Assembly Line Based on ARMA-ANN Modeling

Annamalai Pandian (Saginaw Valley State University, University Center, MI, USA) and Ahad Ali (A. Leon Linton Department of Mechanical Engineering, Lawrence Technological University, Southfield, MI, USA)
Copyright: © 2014 |Pages: 18
DOI: 10.4018/IJAIE.2014070102
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

This research paper aims to predict the automotive Body-In-White (BIW) robotic welding assembly line performance. A combinational prediction model based on the Autoregressive Moving Average (ARMA) and Artificial Neural Networks (ANN) is developed. Classical methods are often used to predict the assembly line throughput, but not ideal. A combinational prediction model is applied for comprehensive analysis and prediction of the assembly line throughput. The various case studies presented in this paper indicate that the precision of the model is better than the other models. This research has significant practical value to the assembly plant because, based on the prediction, plant can make commitment to achieve the production to meet the market demand. Unpredictable performance of the assembly line in the plant leads to more overtime, less time for maintenance and eventually hurting the company bottom line.
Article Preview

Introduction

The overall automotive assembly lines consist of three categories: Body shop, Paint shop and Trim, Chassis & Final assembly housed in three separate buildings: In the body shop, various functional robots such as, material handling, welding, sealing and assembly welding robots build the sheet metal body called, BIW. The sheet metal bodies are coated with several layers of anti-corrosion compounds and painted by paint robots in the paint shop. In Trim, Chassis, Final (TCF) area, the painted bodies are trimmed with carpet, engine and transmission systems are hooked up, electronic systems and mirrors are added to the body. Finally, every car is tested in the plant before shipping to the customer. For this study, only the BIW assembly process lines are considered. The BIW is built by various process robots, which are seamlessly integrated with automated man-machine systems. The welding assembly process systems are very complex in nature and are highly automated with minimum human intervention.

The cross-functional team consists of Original Equipment Manufacturers (OEM), advanced manufacturing engineers and plant personnel would perform the Failure Mode, and Effects Analysis (FMEA) for the assembly processes. The current norm is that the tooling suppliers would deliver the tooling and process to the plant with recommended spare parts list based on the plant historical data. The spare parts list is very general in nature and it is inadequate since, there is no hard data to show which sub-assembly line would go down and at what time causing daily production losses. The equipment shipped by the OEM as production ready to the plant may not be production ready. There is not enough time, or resources, or even motivation for the supplier to perform thorough testing before shipping the tooling to the plant.

With labor reduction and funding reduction forced on the plant every year, the plant is struggling to keep up with the daily production goal. The plant personnel may have no idea when a particular process machine would fail. The unplanned and unpredictable sudden machine failures cause over time for the assembly workers causing fatigue, leaving no time for regular system maintenance, and causing more breakdowns. This is a vicious cycle, which cut into companies’ bottom line. By accurately predicting the future throughput based on the previous performance data, the plant would be able to schedule resources upfront to achieve the production goals. This research paper was intend to study XYZ automotive assembly plant performance data for several months and recommend a prediction model based on ARMA-ANN methodology to accurately predict the future daily, weekly or monthly throughput.

The assembly plants spend thousands of dollars on routine maintenance based on general recommendations, but the unpredictable breakdowns cause strain in the system throughput. There is no accurate tool available for the plant personnel to predict the daily throughput. In addition, in this paper, various robotic assembly line processes and its complexities are analysed. The time series prediction modeling using ARMA model and ANN model is found in literatures for many applications such as, epidemiology, economy, or environmental sciences. However, there is not enough work done to predict especially, the automotive assembly line throughput. In this case study, the actual plant data were statistically analysed and validated for the viable model prediction. This paper attempt to predict the daily assembly line throughput based on the ARMA-ANN modeling.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 4: 2 Issues (2017)
Volume 3: 2 Issues (2016)
Volume 2: 2 Issues (2014)
Volume 1: 2 Issues (2012)
View Complete Journal Contents Listing