Predictive maintenance is a method of preventing asset failure by analyzing production data to identify patterns and predict issues before they happen.
Until now, factory managers and machine operators carried out scheduled maintenance and regularly repaired machine parts to prevent downtime. In addition to consuming unnecessary resources and driving productivity losses, half of all preventive maintenance activities are ineffective. The predictive maintenance has quickly emerged as a leading Industry 4.0 use case for manufacturers and asset managers. Implementing modern industrial technologies to monitor asset health, optimize maintenance schedules, and gaining real-time alerts to operational risks, allows manufacturers to lower service costs, maximize uptime, and improve production throughput.
For predictive maintenance to be carried out on an industrial asset, the following base components are required:
- Sensors – data-collecting sensors installed in the physical product or machine
- Data communication – the communication system that allows data to securely flow between the monitored asset and the central data store
- Central data store – the central data hub in which asset data (from OT systems), and business data (from IT systems) are stored, processed and analyzed; either on-premise or on-cloud
- Predictive analytics – predictive analytics algorithms applied to the aggregated data to recognize patterns and generate insights in the form of dashboards and alerts
- Root cause analysis – data analysis tools used by maintenance and process engineers to investigate the insights and determine the corrective action to be performed
Production asset data is streamed from the sensors to a central repository using industrial communication protocols and gateways. Business data from ERP and MES systems, together with manufacturing process flows, are integrated into the central data repository to provide context to the production asset data. Then, predictive analytics algorithms are applied to provide insights for reducing downtime, which are investigated using root cause analysis software.
The benefits of predictive maintenance
Manufacturers and their customers get a range of business benefits from predictive maintenance. The advantages of PdM include:
- Reduced maintenance time– Automatic reports for strategic maintenance scheduling and proactive repairs alone reduces maintenance time by 20–50 percent and decreases overall maintenance costs by 5–10 percent. These insights save the manufacturer and their customers time and money.
- Increased efficiency– analytics-driven insights improve OEE (overall equipment effectiveness) by reducing unnecessary maintenance, extend asset life and enable root cause analysis of a system to uncover issues ahead of failure.
- New revenue streams– Manufacturers can monetize industrial predictive maintenance by offering analytics-driven services for their customers, including PdM dashboards, optimized maintenance schedules, or a technician dispatch service before parts need replacement. The ability to provide digital services to customers based on data presents an opportunity for recurring revenue streams and a new growth engine for companies.
- Improved customer satisfaction– Send customers automated alerts when parts need to be replaced and suggest timely maintenance services to boost satisfaction and provide a greater measure of predictability.
- Competitive advantage– Predictive maintenance strengthens company branding and value to customers, differentiating their products from the competition and allowing them to provide continuous benefit in-market.
Predictive Maintenance with AI
Industrial artificial intelligence can be applied to predictive maintenance and many other use cases in the manufacturing industry, and although we are just in the beginning of exploiting this technology, there are already many facilities benefiting from industrial AI.
AI is perfectly suited to predictive maintenance. It offers a host of techniques to analyze the huge amounts of data collected from the manufacturing process, and deliver actionable insights to reach and sustain manufacturing excellence. These techniques are referred to as Machine Learning algorithms.
Applying Machine Learning to predict asset failure
Predictive maintenance with machine learning looks at large sets of historical or test data, combined with tailored machine-learning (ML) algorithms, to run different scenarios and predict what will go wrong, and when.
Predictive Maintenance ML Algorithms
Advanced AI algorithms learn a machine’s normal data behavior and use this as a baseline to identify and alert to deviations in real-time.
The algorithms required for machine learning must analyze input (historical or a training set of data) and output data (the desired result). A machine monitoring system includes input on a range of factors from temperature to pressure and engine speed. The output is the variable in question – a warning of a future system or part failure. The system will then be able to predict when a breakdown is likely to occur.
There are two main approaches to AI and machine learning for predictive analytics – supervised and unsupervised machine learning – each is relevant for a different scenario and depends on the availability of sufficient historical training data and the frequency of asset failure.