Leveling Up Condition-Based Maintenance With AI, Machine Learning
Seasoned manufacturing professionals know the assets that make their products – the assembly robots, CNC machines, and conveyor belts – are just as important as the final products themselves. These assets have an outsized impact on business success and failure: Just one defect, or a short window of downtime, can exact a steep toll.
Despite this, many manufacturers do not invest in cutting-edge asset management techniques and innovation. When practicing condition-based maintenance (CBM), for example, manufacturers may not be harnessing all the available data, leaving insights and productivity on the table.
It’s not surprising why: CBM in an era of data can be overwhelming. Thousands of machines sharing information in real-time – how to capture it all, never mind make sense of it?
With AI, however, manufacturers can enhance their CBM efforts with every bit of data coming from the factory floor and significantly decrease equipment failure.
Condition-Based Maintenance (CBM)
Most manufacturers practice CBM as part of their broader asset management strategy, even if they’re unfamiliar with the term.
CBM is a preventative maintenance approach centered on closely monitoring assets and determining when repairs are necessary. The discipline has evolved markedly in recent years with the introduction of IoT technologies like remote sensors, which collect key metrics like temperature, vibration, current, torque and humidity. This reduces the need for older maintenance practices like calendar-based, predetermined inspections or run-to-failure. The results can extend the useful life of the asset, reduce maintenance costs and mitigate the risk of failures and down time.
CBM and Machine Learning
CBM data is far more valuable when coupled with machine learning, a form of AI adept at anomaly detection and catching asset issues prior to failure. Technicians cannot realistically parse all this data alone – there is too much information and not enough time.
CBM entails measuring heat, vibration and other asset qualities. Processes like infrared thermography can capture real-time temperatures of motors, bearings, liquid and other asset components. Vibration monitoring can monitor the vibration frequencies of pumps and motors. And oil analysis can monitor the viscosity and acidity of an engine or gearbox’s lubrication. Ultrasonic, pressure and electrical analysis are other common CBM techniques.
CBM without AI might simply alert technicians when an asset is too hot or vibrating irregularly. Machine learning takes this to another level, mining insights from much smaller fluctuations. AI can help find problems with motors, bearings and other components well before they overheat, get misaligned or run dry.
CBM and Generative AI
Generative AI – AI systems that can create original content – are a newer form of AI technology. Like machine learning, generative AI can be folded into CBM approaches, this time for productivity gains.
A crucial part of CBM is the maintenance itself. Technicians need to roll up their sleeves to realign gears, patch air compressors and mend faulty circuits. Historically, this process has been bogged down by paperwork. Technicians must diagnose the precise issue, research failure codes, create and fill out work orders and then guide all this across the finish line.
Generative AI can dramatically reduce the time spent on these next-action processes. AI systems trained on past maintenance data can generate failure code recommendations alongside a confidence score, potentially enabling technicians to diagnose machine issues in moments. The data required to train these models isn’t prohibitive – in some cases, AI systems can be trained with just a handful of inputs. Generative AI can also summarize lengthy or technical documents, and agentic AI systems can shepherd work orders through complicated approval processes.
Generative AI systems can also be leveraged for failure mode effects analysis (FMEA). These systems are purpose-built for specific asset classes, providing customized, automated and intelligent analysis of an asset and each of its components. The AI can be a powerful partner to reliability engineers and shorten their projects from months to days.
CBM and Agentic AI
An emerging innovation is "agentic AI," a type of generative AI that can perform tasks autonomously across complex workflows in a proactive manner. Agentic AI goes beyond machine learning anomaly detection and doesn't need prompting by users like most GenAI chat bots. Instead, it is designed to run independently, constantly analyzing complex workflows across and between asset classes to generate proactive insights and make targeted recommendations. These AI agents can even take action directly, under the authorization of humans.
The potential of AI for CBM isn’t theoretical. At my company, IBM, we worked with Swedish brewing giant Spendrups Bryggeri and their 80-person maintenance team to couple AI with CBM and unlock significantly more efficient and cost-effective maintenance practices. course, CBM is just one part of a broader asset management strategy one that couples AI With reliability-centered maintenance, predictive maintenance and asset lifecycle management at large.
