Predictive Maintenance: Can machines foretell their lifetime?

One of our recent posts was devoted to the on-going digital transformation of the enterprise, which includes the digitization of industrial processes. Maintenance of machines and equipment is a prominent example of this on-going digitization. Nowadays, most industrial organizations employ preventive maintenance processes for their equipment in order to avoid failures that can have catastrophic consequences on the quality and continuity of their operations. Preventive maintenance is about proactively undertaking repair or service processes at regular intervals before a machine breaks down. Despite their salient characteristics, preventive maintenance processes are far from being optimal, as they usually lead to maintenance much earlier than actually required. As a result, most industries are still far from getting the best Overall Equipment Efficiency (OEE) for their machines and equipment, which incur unnecessary disruption of operations along with additional maintenance effort and cost.   In order to alleviate these inefficiencies, industrial organizations are increasingly turning to the next generation of maintenance processes, namely predictive maintenance. Predictive maintenance is about accurately forecasting when machines are likely to start malfunctioning or even fail, as a means to carrying out the maintenance, at the most appropriate point in time that optimizes effort and costs. It’s an entirely new paradigm based on a machine’s prophesy: Machines are able to foretell their remaining lifetime and indicate the best time for carrying out the costly maintenance process.  This “prophesy” is currently considered as one of the killer apps for the digital transformation of industrial processes.

The Machine’s Prophesy: A Data Intensive Problem

How can machines tell their future? They can do this by collecting and analyzing large amounts of data that provide indications about their current and future state. In most cases, such indicators can be provided by sensor data, such as:

• Vibration sensors: We have felt the vibration of a machine that we operate and we know that it can be indicative of mechanical problems. This holds true not only for the engine of our car and our washing machine, but also for machines and equipment in a plant.
• Acoustic sensors: We know that an abnormal sound in the operation of a machine is a signal of potential problems. Collecting acoustic data can be another good idea for predictive maintenance.
• Ultrasonic sensors: Beyond acoustics perceived by humans, it is also possible to detect the status of a machine through ultrasonic sensors that capture sound patters, which are not audible to humans.
• Temperature sensors: Malfunctioning equipment is likely to have increased temperature, which is therefore another maintenance-related indicator.
• Power consumption sensing: Changes in the patterns of a machine’s energy consumption can also be indicative of problems and could therefore contribute to predicting its current or future state.
• Thermal cameras: Thermal images about an equipment can also be used to indicate its status, as they are affected by increased heat of the electrical circuits of the equipment.

In addition to sensor data, other data sources can be used such as oil analysis data stemming from the oil condition and the inspection of the lubrication system, as well as quality data relating to the machines operation. The latter are important, given that any malfunction of a machine reflects directly on the quality outcome of the industrial processes.

Based on one or (usually) more of the above-listed datasets, predictive maintenance systems can apply predictive analytics in order to forecast the future state of the machines. The outcomes of predictive analysis should not be limited to producing a prediction about the time when a machine will fail. Rather it should solve a more complex risk management problem i.e. to identify the optimal time for maintenance. The latter time is the one that optimizes OEE and minimizes break-down risk at the same time.

Predictive analytics for maintenance are usually based on emerging Artificial Intelligence (AI) algorithms for data processing, which are based on neural networks, process multimedia data and are known as deep learning. In essence, predictive maintenance is a Big Data problem as a result of the need to process very large volumes of data streams from highly heterogeneous sources.

Predictive Maintenance Components

From an IT perspective, a predictive maintenance system comprises of the following components:

• An Internet-of-Things (IoT) based infrastructure for data collection from multiple sensors for equipment condition monitoring and other business information systems such as Enterprise Resource Planning (ERP) and Asset Management (AM) systems. Based on this infrastructure, data that were previously fragmented are integrated and used for predictive analysis in a unified way.
• AI based data analytics components for predictive analysis, which enable the fast processing of diverse data sources (including multimedia data) towards deriving predictive insights about the health of assets.
• Data reporting and visualization systems enabling the ergonomic and user-friendly visualization of the predictions, while at the same time facilitating access to the individual datasets and predictive insights that produce them.
• An advanced cyber-security infrastructure, which protects datasets, while at the same ensuring the trustworthiness of data-driven interactions between the different systems.

Consumer Aspects

While the industry is leading the predictive maintenance revolution, we will soon see such capabilities integrated in consumer products, such as our cars. The latter will come with maintenance-as-a-service capabilities i.e. the ability to carry out the maintenance, repairs and service of the vehicle at the best possible time towards ensuring optimal economic efficiency for the customer. Such revolutionary approaches will provide tangible economic value for consumers, while at the same time enabling a radical shift in the business models of the automotive vendors (e.g., new wave of after sale services).  BMW, the German automotive manufacturer, is already working in this direction.

As machines become oracles of their fortunes, we can expect the world to be a little more predictable.