Predicting field performance of on-board diagnostics using statistical methods

On-Board Diagnostics will play a crucial role in the emerging era of the Internet of Things. With billions of devices deployed, traditional manual preventive maintenance approaches will be cost prohibitive. As we make these small autonomous devices intelligent, it is critical that we also give them a very advanced ability to assess and report their own health. Of course, on-board diagnostics are not a new concept. What is new with the Internet of Things is the extreme economic leverage that the on-board diagnostics will have due to the huge number of devices to be developed. If the on-board diagnostics perform poorly, the resulting surge of support costs could be enough to drive the organization deploying the Internet of Things systems out of business. As an organization gets ready to release millions of a certain type of Internet of Things device into the wild, how confident can the organization be that the on-board diagnostics will really perform as expected? As it turns out, we can learn from history. In the 1960s and 1970s the mainframe and telecommunications industries developed powerful statistical methods for answering this exact question for the large mission-critical systems that they were deploying. This paper will review these historical approaches and show how they can be practically applied to a wide variety of diagnostic systems. The paper will show: How to set business performance objectives for the on-board diagnostics · Practical methods of quantifying failure modes · The statistical measurement approach · How to prepare testing programs · How to administer the tests and evaluate the results.