Predictive Maintenance Applications and Anomaly Detection

Ali Özgür (Writer) 13 March 2024

Among the 16 major losses taken into account in the calculation of the OEE value, which is the most important indicator of the production efficiency of the enterprises in the manufacturing industry, the most important category is the 8 major losses under the title of Equipment Effectiveness. Among these 8 losses, breakdown losses can be reduced directly and defective product losses can be reduced indirectly by predictive maintenance practices and OEE values can be improved.

Predictive maintenance in the manufacturing industry is a proactive approach that aims to plan equipment maintenance by predicting the failures that may occur in the equipment used in production. Predictive maintenance applications cover all stages that we can summarize as defining equipment specifications, defining threshold values, collecting data from the field, enriching data, predicting and triggering maintenance actions. Predictive maintenance can be implemented using software systems, manual processes or both methods at the same time.

Traditional predictive maintenance approaches are based on following the limit values defined by equipment manufacturers and valid under normal operating conditions by matching production plans and equipment used in production. One of the difficulties of the traditional approach in practice arises from the fact that the equipment is used in more challenging conditions than normal conditions due to production planning or physical conditions. Accordingly, the limit values set by the equipment manufacturer change dynamically and the need for maintenance arises at different times in each cycle. In order for this method to produce healthy results, equipment usage must be monitored regularly.

Organizations that want to make more accurate predictive maintenance estimates prefer to use applications that can directly read values recorded on equipment or collect data from equipment equipped with external sensors and the environmental conditions in which the equipment is operating. Equipment readings provided by signaling and recorded over long periods of time create a data stack of millions of data points at the end of the day. By correlating this big data on equipment values with other inputs such as downtime and planned maintenance, predictive maintenance applications can be performed with methods that aim to predict when the next maintenance should be performed. The most important disadvantages of this method are that the cost of collecting and recording big data is high, correlation of equipment values with other data collected from the field is a labor-intensive process, the forecasting algorithms used are not sensitive to instantaneous changes in operating conditions, and most importantly, generating forecasts with this method takes a very long time as the sensitivity of the algorithms used increases.

In most equipment failures, abnormal signals start to be generated in the equipment or in the environment in which the equipment operates before the failure. Continuous monitoring of these signals and detection of abnormal values form the basis of next generation predictive maintenance applications. In this method, maintenance prediction can be made by analyzing limited data sets and taking into account temporary norms in operating conditions, without the need to record signals for long periods of time.

trex MES platform In order for companies aiming to perform predictive maintenance to make healthy decisions, they must first be able to accurately calculate OEE values, analyze the details of equipment losses well, and quickly examine other indirect losses such as equipment-related quality losses. trex MES platform provides ready-made analyses that will enable manufacturers to make healthy decisions while planning their investments in predictive maintenance. With this platform, we offer our customers identification for predictive maintenance applications, data collection from different sources, association of equipment data with other production data, analysis on big data or instant limited data sets, rich library of anomaly detection and trend analysis algorithms, rich action options (such as creating maintenance work orders, sending sms, e-mail, mobile notifications).

As in predictive maintenance, we pay attention to developing innovative and pioneering technologies and services that will provide significant benefits to companies and prevent potential problems they may experience, increase their profitability and keep them ahead of the competition.