Introduction

Manufacturers normally follow a schedule of preventative maintenance (PM) on their equipment. This requires that the number of hours a piece of equipment is running is tracked. Unexpected downtime increases the cost of manufacturing. Things such as lost run time, employee downtime, and repair and diagnostics costs and time directly influence the cost to create the product. Maintenance tasks that include everything from fluid replacement to parts replacement are scheduled based on those hours. There are times when something may wear faster than expected, or there might be an anomaly in the equipment. Predictive Maintenance assists in identifying possible issues pre-emptively.

Objectives

Manufacturing Perspective

The integration of predictive maintenance in manufacturing environments can allow manufactures to address potential issues proactively before they lead to costly downtime or failures. Predictive maintenance can prevent downtime (improve availability), extend the lifespan of equipment (improve performance), and ensure product quality (improve yield),  all of which are essential for optimizing OEE.

Predictive maintenance leverages data analysis to forecast when equipment may fail or need servicing. By continuously monitoring machinery's condition and performance, manufacturers can schedule maintenance ahead of time, preventing unexpected breakdowns that could halt production.

With data flowing from various machine performance output or sensors, a machine can be monitored for conditions that indicate that a failure is coming. For example, if a motor is drawing more electrical current than it normally draws, that motor is working harder than normal. This could be an indication of a motor failure or that the motor drives are not moving as freely as before, indicating that it needs attention. Replacing a malfunctioning part can speed production back up after a slowdown, or prevent unscheduled downtime for maintenance by replacing an element before it fails. 

Not every anomaly is an outlier. It is important to remember that defining an outlier isn't a subjective process. There's also noise to factor into the analysis. It can be difficult to assess the difference between normal data, noise, and the actual anomalies. 

Some solutions are obvious with a particular anomaly arises, while others require more research. It is the part of the data analyst to identify these anomalies as quickly as possible to forestall expensive shutdowns or higher numbers of product that do not pass quality standards.

 

Data Continuum: From Normal to Anomalies

A visual continuum from left to right showing "Normal Data," "Noise," and "Anomalies" as segments increasing in red hue and pointing right. The graphic illustrates that data becomes increasingly outlier-like from normal to anomalous, with noise and anomalies classified as weak or strong outliers.

Image by Tim von Hahn, inspired from Charu C. Aggarwal in Outlier Analysis

 

Key Data Categories

Consider the following conditions and consider what anomalies in these parts or systems could indicate.

 

Note
Different models can be used to calculate maintenance so that repair can be scheduled.