Production testing of spark plugs using a neural network

Despite nearly 150 years' evolution, there have been relatively few advances in the design, and methods of production testing, of spark plugs. For years, an ingenious yet relatively simple "go/no go" batch test has been favoured, yet this testing solution exhibits some major disadvantages. This paper describes an alternative method of spark plug testing, offering elementary diagnosis of faults as well as detection. In this functional test regime, spark voltage waveforms are classified using a neural network. The promising results of this experimental work indicate that neural networks may offer considerable potential for the future of spark plug testing.