So much for the theory… Ford patterns elusive

A recent study into the feasibility of using data mining techniques to predict when machines are likely to break down has produced inconclusive results. The study, started at the beginning of this year, involved Ford, the University of East Anglia and Lanner Group, a software developer. It was funded by Ford. `Machine breakdowns happen more […]

A recent study into the feasibility of using data mining techniques to predict when machines are likely to break down has produced inconclusive results.

The study, started at the beginning of this year, involved Ford, the University of East Anglia and Lanner Group, a software developer. It was funded by Ford.

`Machine breakdowns happen more frequently than we would like and we wanted to see if data mining could establish patterns in machine failures,’ says John Ladbrook, European simulation technical specialist for Ford’s powertrain operations. `It was a new way of looking at our data.’

Ford already collects data from many of the machines in its factories. On its most recent lines, every machine is automatically monitored 24 hours-a-day. The monitoring system records the times when the machine is operating normally, when tools are being changed and when breakdowns occur.

By applying data mining, Ford and its partners hoped to identify patterns in the data. For example, it might have shown that a minor breakdown of one machine was regularly leading to a more serious breakdown of another.

For the feasibility study, a month’s data from Ford’s Spanish powertrain plant was used. Patterns were identified in the data but Ladbrook questions their usefulness.

`When machines break down they are repaired, which alters the patterns, so we would need to analyse the data on a regular basis to identify new patterns,’ he says. The useful life of a pattern, therefore, would be limited. `With data mining, I think you need reasonably stable conditions,’ he adds.

While the feasibility of using data mining to predict machine behaviour has not been proven to Ford’s satisfaction, the company has used other techniques to model breakdowns to predict behaviour of future production lines. This work was carried out by Ladbrook, using manufacturing process simulation technology.

`I used historical data about the breakdowns that occur on our existing powertrain plants and incorporated that into a model of a future line to try to predict how that line will behave,’ Ladbrook explains.

The main software used was Witness, also from Lanner, which is widely used by Ford. This software allows users to construct a computer model of a manufacturing process, simulate its functions and analyse its performance.