Machine-Learning and Anomaly Detection: Stop Problems Before They Start
Part Two: Why Anomaly Detection is the Natural Preceding Step Before Predictive Maintenance
Anomaly detection, also known as outlier detection, occurs through the process of data mining. It often includes identifying events or items that do not mold to meet expected patterns in current datasets. These events can result in serious technical errors such as structural problems or even bank fraud.
Predictive maintenance is a process designed to examine the functionality and workings of in-service equipment. In looking at these functions, decisions can be made regarding what maintenance should be performed and when, and how equipment should be handled. Executing this process usually allows businesses and industries to save money by performing maintenance tasks only when they’re needed instead of on a routine or preventative basis.
YOU CAN’T HAVE ONE WITHOUT THE OTHER
As you can see, predictive maintenance cannot be fully executed unless anomaly detection occurs first, primarily because maintenance without accurately checking for problems is a waste of time and money. However, understanding these anomalies before they strike can be particularly hard, especially because most of the time it’s difficult to pinpoint what to look for or find. What should one keep an eye out for? It’s not always easy to tell, especially when you’ve never seen it before.
SEE ANYTHING… STRANGE?
The basic idea is to watch for anything that appears out of the ordinary. As basic and cliché as this sounds, it’s a solid fact. If any disruption occurs in the daily patterns of a working operation, it’s best to sound the alarm early and check things out before they get any worse and the system has moved in an opposite or unworkable direction.
HOW TO HANDLE MAINTENANCE
The most common method for detecting anomalies early appears to be what’s known as time series analysis, which involves studying and recording historical data of current and/or previous operations. Items like time and frequencies are standardized, and dates are pinned to the action being recorded. The data produced is then able to show executives and technical teams just how “normal functioning” is supposed to look. Hence, in examining future data, inspectors are inherently able to identify anything that does not match or appear “normal.” These early signs of disruption can thereby be tackled early before they interfere with operations or become too menacing and large to fix.
Another popular (though relatively new) method of detecting anomalies is known as the iFOREST algorithm. It works to isolate observations by “randomly selecting a feature and then selecting a split value between the maximum and minimum values of the feature.” This method is sometimes considered easier and more feasible as in general terms, only a few conditions require consideration when separating defective cases from normal ones. However, the number of conditions necessary can sometimes vary depending on the case in question.
THINK BIGGER, SAVE MORE
Naturally, maintenance can be costly, but not finding or dealing with problems beforehand tends to run up even higher expenses, which is why anomaly detection is considered so important. Addressing deficiencies during their early stages ultimately paves the way for predictive maintenance to take full effect. Furthermore, it’s slated to save factories and businesses a whopping $630 billion by the year 2025.
Regularly monitoring data is likely to nail problems on the head before they take on even larger forms. Diagnoses can then be made to fix these issues and keep equipment working throughout the duration of its life while also reducing costs and the overall downtime for a business.
Silipo, Rosaria. “Anomaly Detection in Predictive Maintenance with Time Series Analysis.” KDnuggets Analytics Big Data mining and Data Science. KDnuggets. Web. 17 Apr. 2017.
“Predictive Maintenance: What’s the Economic Value?” Anodot. Anodot, Inc., 14 Mar. 2016. Web. 17 Apr. 2017.
Winters, Phil, Iris Adae, and Rosaria Silipo. “Anomaly Detection in Predictive Maintenance.” Knime.com. Knime.com, Web. 17 Apr. 2017.
Bahnsen, Alejandro Correa. “Benefits of Anomaly Detection Using Isolation Forests.” Easy Solutions. Easy Solutions, Inc., 02 Nov. 2016. Web. 22 Apr. 2017.