Significant time and cost benefits are available now to enhance smart maintenance regimes, using prevention rather than cure. The near future is even brighter, where the roll-out of Industry 4.0 technologies are enabling companies to connect and visualise machine data. Component life expectancy data can be combined with cloud-hosted application data. Bringing in the power of artificial intelligence and machine-to-machine learning will further dramatically change the possibility to predict and prevent failures and improve OEE.
While machine downtime has obvious financial costs, the reputational damage that comes with it can have an even greater impact. Take the example of the automotive sector. In March this year, Tesla recalled 123,000 of its Model S cars over a faulty steering component. Shares in the company fell 4% immediately after the announcement, costing the company millions of pounds and untold levels of reputational damage.
“By deploying an integrated cognitive and machine-first technology that runs end-to-end in the manufacturing and post-purchase lifecycle, manufacturers can ensure that these processes run like clockwork,” says Ruban Phukan, product vice president at software and data business Progress. “In general, cognitive predictive maintenance can reduce breakdowns by 70% and reduce maintenance costs by 30%.”
Also, being able to predict problems ahead of time helps organisations to achieve higher productivity by planning resources and inventory better and significantly reduce scraps and rework.
“I think it’s safe to say that the future of maintenance is in embracing data and digital technologies,” Deborah Sherry, senior vice president and chief commercial officer at GE Digital Europe, Russia and CIS, says. “AI and advanced data analytics are being increasingly used in the manufacturing sector to improve operational efficiency by monitoring industrial assets and using this data to drive predictive maintenance.
“Advanced data analytics and AI, combined with asset performance management applications, can help organisations to identify issues before they have occurred. This predictive maintenance approach is most often used to repair critical equipment and systems (those whose failure or malfunction would be so costly that they must be avoided at all costs) before a breakdown happens.
“AI can also be used to facilitate complex system scheduling and management to better control the parameters of processes that include multiple machines and sources of uncertainty. Those sources may relate to the variable quality of inputs, to weather conditions (the impact of air temperature or humidity on a chemical process) or human intervention.”
Rafi Billurcu, a partner for manufacturing at Infosys Consulting, believes that no one in industry can ignore the ubiquity of the Internet of Things (IoT). He adds that embedding intelligent IoT technologies into machines promises to transform run-of-the-mill manufacturing facilities into smart factories. “The impact will be seen across all aspects of operations from asset management and machine maintenance to planning, quality control, and even field service,” he says. “The introduction of smart sensors, one such IoT technology, promises to enable the real-time monitoring of machines – but predictive maintenance is not possible with sensors alone.”
Searching for extra intelligence
The consensus among the experts is that the key to realising the full potential of IoT is to employ artificial intelligence technology alongside it. “In order to put the vast amounts of data generated by these IoT sensors to good use, organisations must invest in AI as well – or risk failing to make predictive maintenance worthwhile.
“Using AI means maintenance can be predicted based on previous machine data, mapping it out across a longer period of time and spotting anomalies and changes that an average human couldn’t,” Billurcu adds. “What’s more, integrating machine learning into a system’s analytics capabilities can increase the accuracy of the predictive algorithms, enabling a more sustainable system that can learn over time, offering an even better return on investment. Although there is a high upfront cost, employing these technologies will mean engineers can avoid unplanned downtime and minimise negative consequences.
“Applied across the entire supply chain, this technology could save millions. Run-to-failure won’t cut it in the smart factory of the future – it doesn’t make business sense.
“Predictive maintenance, backed up by AI technology, holds the key to streamlined supply chains that are imperative for success in Industry 4.0.”
Another interesting trend that Sherry highlights in machine maintenance in the manufacturing sector is the growing use of digital twins or digital replicas of industrial assets to test and simulate different performance scenarios, operation optimisations or new business models virtually, without putting operations at risk. Digital twins can help companies to uncover deep patterns of behaviour and get the most out of each asset by integrating analytics from digital twins across an entire class of assets. A combination of AI and ‘old-fashioned’ domain expertise can help companies to predict potential issues – and make sure that they don’t happen.
“In the future, every machine will have a digital twin with the ability to connect a system, or systems, of digital twins easily,” Sherry adds. “As more digital twins are created and connected to a digital platform, the industrial learning system will be able to feed data back to the individual digital twins, improving fidelity. This will drive greater productivity and efficiency and will significantly improve systems maintenance.”
Phukan is convinced that digital twins have changed the way manufacturers should be looking at maintenance. “Pairing machines with a digital double that mimics actions and environments, as well as the same working conditions and depletion, provides a wealth of real-time data that helps manufacturers to understand the health and readiness of equipment down to the molecular level,” he says. “This is all possible because of the advancements in sensor technology that generates the data and AI and machine learning that help engineers to automatically process the data to detect and address the tiniest of issues that would otherwise be missed during a manual inspection process.”
Cognitive anomaly detection
Organisations often focus on historical failure and look to identify patterns that will help alert them when past failures or faults are primed to reoccur. Although this is very valuable, they need to keep in mind that only about 20% of the problems that occur in the field are in fact repeat problems. This means that 80% of the problems and faults are new and unknowns for most industries. Unless the signs for these 80% of the problems are detected early on, modelling on past failure patterns won’t bring the desired results. Cognitive applications, on the other hand, are able to teach themselves continuously and go beyond the macro-patterns that the human brain tends to spot.
AI and machine learning can baseline the various different normal conditions and also continuously adapt to changing conditions. “This then helps identify anomalies which deviate from the baselines that indicate a future problem and initiate preventative action, sometimes even automatically,” Phukan explains. “They can, in fact, do that individually for each and every asset even at a great scale.”
Phukan believes that data is the new oil. But raw data is like crude oil – it lacks use unless it goes through a complex data refinery process to generate insights that can propel any organisation forward to significant business gains. “True predictive maintenance can only be achieved if all of the data collected from the machines can be processed in a timely manner, at scale to identify the tiny hidden signals,” he says. “What makes this process challenging is that most failures and problems that occur are rare events and they are scattered over the massive volumes of Industrial IoT data that high frequency sensors can generate. The other challenge with the raw sensor data is that they are unlabelled, meaning it is difficult to know whether a particular value of a sensor is a good value or a bad value unless it is put in the context of the operating conditions and also other sensors.”
The data refinery process to transform raw data into intelligent information involves processing sensor-level information for each and every machine individually, baselining for the normal at a sensor level and then combining sensor characteristics to generate baselines
at the machine level, adapt to changes in environmental and operating conditions, identify the deviations and match signals with failure patterns. “All these make it extremely difficult to do it manually,” he continues. “The cognitive approach can help automate these complex steps at production scale to deliver intelligent information to make decisions.”
The future now
Within its own production facilities, Festo uses Near Field Communications (NFC) for digital maintenance support. This means that machines within the production plants can present data directly to the maintenance team without passing through a central database.
“For the maintenance team, this offers new ways to interact with the factory equipment and gather useful data,” says Steve Sands, head of product management, Festo. “Their main tool is a tablet equipped with a custom-developed app which, together with a mobile depth sensor, enables the user to access more information the closer they are to the components requiring attention.
“At a high level the tablet may show an overview graphic of the entire plant layout and indicate that there is an alert raised on a machine on a production line. These alerts follow a simple grading system to indicate their urgency, enabling the team to prioritise their workload effectively. This makes the team far more responsive because they can direct their resources to the highest priority notifications first, such as those where a pending machine stoppage would seriously affect planned production.
“For less serious events, information about the reasons for the alert can be interrogated remotely without any need to interrupt production. The lowest category alert will simply be a notification that a part or a peripheral (such as ink in a label printer) requires replacement on a given date.”
When intervention is necessary, co-ordination of people and parts is much slicker in the automated factory. The smart factory maintenance team can check the availability of components and tools so that everything necessary to undertake a scheduled maintenance task
is to hand where and when it is needed. “Using their tablets as an interface, the engineers can access information directly from the machine and its component parts, call up maintenance instructions, consult detailed 3D models of the machine to check aspects like accessibility prior to starting work and even access detailed technical manuals or online help direct from equipment suppliers during the task,” Sands concludes.
End game for maintenance
The end game is to achieve autonomous maintenance where the machines maintain themselves and seek expert intervention only when necessary. Cognitive algorithms can identify issues well ahead of time, initiate corrective actions or schedule maintenance based on the optimal time that has the lowest impact. It can check inventory and automatically place orders for replacement parts with the vendors which can fulfil the request in time at the best price. It can schedule time with the nearest available field engineer with the right experience, presenting them with the maintenance plan. All that the engineer is then required to do is follow the steps.
“With the advancement of robotics, the field engineer can be a robot trained to follow the maintenance steps, which further minimises active human intervention,” Phukan explains. “With the convergence of technologies like AR/VR, a human expert can monitor the maintenance activity remotely and even control the robot for certain complex tasks where automated action may not be possible. With the addition of technologies like blockchain, a faulty part can be traced back to its origin and even identify the source of the fault. Also, parts can be tracked effectively as they move through the supply chain which minimises the manual logging effort, and can facilitate efficient routing as well as reducing fraud.”
Predictive maintenance can reach a whole new level with product recalls, missed SLAs and operational inefficiencies becoming a thing of the past. This can deliver unprecedented customer experience, higher margins, lower risks and a complete transformation of how manufacturing industry operates.