Nural Networks Modeling Strategy: a Study on Abrasive Flow Machining Method |
This paper discusses the preliminary development of aneural network-based process monitor and off-line controller for abrasive flowmachining of automotive engine intake manifolds. The process is only observableindirectly, yet the time at which machining achieves the specified air flowrate must be estimated accurately. A neural network model is used to estimatewhen the process has achieved air flow specification so that machining can beterminated. This model uses surrogate process parameters as inputs because ofthe inaccessibility of the product parameter of interest, air flow rate throughthe manifold during processing. The objectives of this research are to improve thefunctional performance of automotive engines and to enable cost effectiveprocess control of the AFM process. A neural network model is used to capturethe nonlinear relationship between the AFM media and the specified outgoing airflow rate by using part characteristics such as surface finish and weight andprocess parameters such as media temperature. This model allows the predictionof when the machining process should be terminated to meet the air flowspecification and can be used as an off-line controller for the process. Abrasive Flow Machining (AFM) is an effective way topolish unsymmetrical surfaces and interior structure of parts, which aredifficult to reach by conventional machining. The material to be machined maybe cylindrical or any complex shape. Various process parameters are MaterialRemoval Rate, Machining Time, and Abrasive Mesh size that affects theperformance of AFM. The objective of this paper is to study the effects ofprocess parameter related to it such as material removal rate, surfacefinishing etc. In recent years, several papers on machining processeshave focused on the use of artificial neural networks for modeling surfaceroughness. Even in such a specific niche of engineering literature, the papersdiffer considerably in terms of how they define network architectures andvalidate results, as well as in their training algorithms, error measures, andthe like. Furthermore, a perusal of the individual papers leaves a researcherwithout a clear, sweeping view of what the field’s cutting edge is. Hence, this work reviews a number of these papers,providing a summary and analysis of the findings. Based on recommendations madeby scholars of neurocomputing and statistics, the review includes a set ofcomparison criteria as well as assesses how the research findings werevalidated. This work also identifies trends in the literature and highlightstheir main differences. Ultimately, this work points to under explored issuesfor future research and shows ways to improve how the results are validated.