There are three types of industrial maintenance: reactive (emergency) maintenance, preventive maintenance, and predictive maintenance. In reactive maintenance, repairs are made only after the equipment fails. Preventive maintenance focuses on conducting inspections at scheduled intervals. Parts that are prone to wear are replaced on a schedule. One downside to this method is that parts are sometimes replaced before the end of their useful life. Additionally, downtime may be increased, because inspections often require equipment to be shutdown for inspections.
Predictive maintenance (PdM) is a conditional maintenance strategy that relies on sensors for remotely monitoring the condition of the equipment. The sensors provide real-time equipment data and computational analytics to predict when maintenance work is needed. Advanced communication technologies, sensors, and analytics make predictive maintenance increasingly practical. PdM also pinpoints the root cause of problems in complex machinery and identifies early component failures, reduces unplanned downtime, maximizes equipment lifetime, and avoids costly repairs.
PdM is enabled by data. The condition of the equipment is monitored in real-time and maintenance is scheduled before failures happen. PdM solutions use artificial intelligence (AI) software to monitor assets and generate alerts when unusual conditions, such as abnormal vibrations or high temperatures, are detected.
To build and deploy predictive maintenance algorithms, the following stages are required:
- access sensor data
- preprocess data
- extract features
- train the model, and
- deploy and integrate the model.
Model training includes the detection of anomalies and the classification of the different types of faults. In this stage, the machine’s remaining useful life (RUL) can also be estimated. The data generated is used to develop an algorithm that is able to describe the system in a range of healthy and faulty conditions. After the algorithm is developed, the software is deployed in the cloud or on edge devices.
Preventive maintenance can miss failure signs in between inspections. Unlike preventive maintenance, predictive maintenance solutions offer 24/7 monitoring, sending out notifications whenever anomalies are detected. Migrating from a preventive maintenance solution to a predictive maintenance solution can potentially reduce system downtime. The migration involves the following steps:
- replicating engineer knowledge for the targeted asset,
- establishing an acceptable conditions baseline,
- retrofitting one or multiple sensors to the target, and
- setting warning and alarm thresholds.
Typical predictive maintenance examples include HVAC (heating, ventilation, and air conditioning), a mission-critical component in healthcare and industrial settings. Experience has shown that HVAC fans, motors, and compressors are the components most susceptible to failure. Information about airflow, current consumption, sound, and vibration can help understand whether the unit is running smoothly or will soon cease to function.
To better understand the difference between preventive and predictive maintenance, let’s consider a bottling factory. A preventive maintenance strategy typically used in bottling factories involves manually inspecting the bottling line motors at three and six-month intervals. With a predictive maintenance solution, software running an AI algorithm would run distortion analysis in real-time. Because the motor vibration, temperature, current, and insulation resistance are constantly monitored, the need for manual inspections is considerably decreased.
