Preventative Maintenance Data Mining Overview
One of the benefits of storing vast amounts of data is the ability to use this data to search for patterns to make your business more efficient. For businesses that rely on equipment that needs regular maintenance data mining can provide an advantage over your competition. Historical data is used to train a data mining model to look for patterns in the data. Two methods leveraged to look for patterns are Regression and Classification techniques.
Preventative Maintenance Examples
Aircraft Example –
The section below is taken from a paper on data mining to help keep aircraft flying at optimal performance.
“For Aircraft Launch and Recovery Equipment (ALRE), the goal is to get planes in the air and ensure they land safely. Consequently, a high operational availability (Ao) is crucial to ALRE operations. In order to ensure high Ao, it is crucial that the amount of maintenance, both corrective and preventative, is kept to a minimum. Historically, improvements have been reactive in nature to satisfy the Fleet’s needs of the moment and are never implemented across the Fleet. One approach to improving maintenance practices is to use historical data in combination with data mining to determine where and how maintenance procedures can be changed or enhanced. For example, if a maintenance manual says to remove three electronics boxes based on a built-in test (BIT) code, but historically, the data shows that removing and replacing two of the boxes never fix the problem, then the maintainer can be directed to first remove and replace the box which the data suggests is the most-likely cause of failure. This type of improvement is where data mining can be used to enhance or modify maintenance procedures.” from IEEE.
Wind Turbines – Example
The example below shows how SCADA data measurements can be used to help determine if a part is at risk of failure.
“Wind turbines use a number of temperature sensors to monitor various components and shut down the wind turbine if the component exceeds an alarm level. Most SCADA historians collect 10-minute minimum, maximum, average and standard deviation analogue values. A temperature sensor will typically fail open or short circuit. When this happens, the measured temperature will be a limit value (e.g. 250 or -50°C). When the sensor begins to fail, it can exhibit short duration excursions to these limit values. Because of the short duration of limit value occurrences, the wind turbine control system does not shut the turbine down (it uses an average temperature to reduce noise susceptibility). These limit values are however, visible in the 10-minute minimum and maximum values and their frequent occurrence always precedes sensor failure as they represent an intermittent open/short circuit.” from SCADAMINER