Bruce David Bryant

Neural network application to comprehensive engine diagnostics

The authors examine the application of trainable classification systems to the problem of diagnosing faults in engines at the manufacturing plant. It is demonstrated how a combination of conventional statistical processing methods and neural networks can be combined to create a classifier system for engine diagnostics. The most significant computational effort is required for the principal component analysis and to properly develop the hard-shell classifiers using data sets augmented with Monte Carlo methods. Once these procedures are carried out, the application of neural networks to the data set to obtain the trainable classifier is quite straightforward.<<ETX>>

Case Example 2: Data analysis for diagnostics and process monitoring of automotive engines

This case example describes software for analysis of data from cold tests of fully and partially assembled engines. Cold testing of engines (testing without fueled running) at engine assembly plants has attracted considerable interest for its potential to reduce test costs and improve product quality. Cold testing of newly manufactured engines offers several advantages over traditional ‘hot’ tests (fueled, running tests): Speed: fewer hookups and detachments, shorter overall test time. Cost: less expensive test and facility hardware; cold testing also offers improved safety and environmental conditions. Quality: extremely sensitive to mechanical deviations which affect performance and therefore much lower probability of missing a failure. Coverage: wide range of potential defects can be detected and identified.