Malfunctioning components and systems in wind turbines, manufacturing plants, aircraft and other sectors are typically addressed by “repair and replace” maintenance activities – an approach that expends both time and money. A more efficient model would be to predict failures so that corrective actions could be implemented before malfunctions occurred.

Predictive maintenance can lead to major cost savings, higher predictability and increased system availability. Source: Radwell InternationalThat’s the premise, and promise, of predictive maintenance: to anticipate failure and launch remedial operations. This can lead to major cost savings, higher predictability and increased system availability. Additional benefits relative to conventional preventive maintenance include minimal downtimes and optimization of periodic maintenance requirements.

The approach relies on data, machine learning, artificial intelligence and modeling to make predictions about future outcomes. Maintenance work is scheduled based on diagnostic evaluations that determine when to perform service. The monitoring of equipment conditions provides trending data to help anticipate future maintenance needs.

Sensors incorporated into machinery, whether a CNC system or an aircraft engine, monitor and collect data about its operations. That data includes a timestamp, a set of simultaneous sensor readings and device identifiers. The goal is to predict at a given time whether equipment will fail in the near future.

Monitoring Tools and Techniques

Numerous tools and techniques are available to monitor the condition of machines and equipment and identify symptoms of wear and other failures:

• Power system assessments take the form of visual inspections of a power distribution system, allowing defects, deficiencies, deteriorations, hazards or weaknesses in existing system installations to be identified.
• Infrared inspections use a specialized camera to detect anomalies not apparent to the naked eye. These assessments are used to identify “hot spots” that can be a precursor to equipment malfunction in an electrical setting: Worn components and malfunctioning electrical circuits generate heat that will show up on the thermal image. Infrared inspections often can pinpoint potential problems before they lead to a breakdown or cause significant damage.
• Online temperature monitoring provides 24/7 access to critical connection points where traditional thermography cannot be used. Wireless temperature sensors can be installed in low-voltage and medium-voltage equipment areas not accessible to infrared cameras, where they will function during planned outages. Continuous monitoring provides the means to evaluate current equipment current conditions and detect abnormalities at an early stage.
• Vibration analysis can be performed with a hand-held device or by monitors built into the machinery. As shafts and bearings begin to wear and fail, they produce different vibration patterns that can be recognized by trained individuals. By comparing readings against known failure modes, these analyses can quickly identify problems and help to pinpoint exactly where problems are occurring.
• Insulating fluid analysis is used in an oil-filled transformer to measure the physical and chemical properties of oil. Common tests performed on electrical insulating oils include readings for moisture content, acid levels, dielectric strength, power factor and dissolved gas analysis. An oil analysis also can detect the breakdown of an oil paper insulating system.
• Circuit monitor analysis helps facility managers and engineers to understand where and when dangerous and destructive transients, sags and swells occur by recording data relating to voltage, current and power.

There are also some noteworthy emerging tools:

• MathWorks recently developed Predictive Maintenance Toolbox, a MATLAB product that helps engineers to design and test algorithms for condition monitoring and predictive maintenance. The tool’s capabilities include data organization and labeling, condition indicator design and remaining useful life (RUL) estimation. Reference examples for motors, gearboxes, batteries and other machines also can be reused for developing custom algorithms.
• Artificial intelligence techniques such as deep learning can be applied to noise pattern assessment of underperforming machines, another tool for predicting problems in advance. Israeli startup 3DSignals employs ultrasonic microphones linked to the company’s internet of things (IoT) service. Data gets processed and uploaded to an online network, where deep learning algorithms take over.
• Remote maintenance via drones is increasingly being adopted to collect data about system conditions. Autonomous aerial systems can monitor issues affecting the operation of pipelines, wind turbines and other outdoor systems without incurring increased labor costs or risks to workers.

Management of data flows generated by sensors and instruments like these is a core component of predictive maintenance programs. There must be technology in place to collect, process, prepare and structure the massive amounts of device data that will be stored within an organization’s ecosystem. This system also must be able to understand what each piece of data represents, so that it can be monitored as a part of the entire maintenance feedback landscape.

According to MarketsandMarkets, the predictive maintenance market is expected to grow from $1,404.3 million in 2016 to $4,904.0 million by 2021, at a compound annual growth rate (CAGR) of 28.4% during the forecast period. The growth is due to rising dependence on big data, emerging technologies such as the IoT and an increased company focus on operational cost reductions.