The challenge: credible predictions from unpredictable data
Most predictive maintenance systems rely on machine learning models that are trained on the behaviour of a specific machine. These models learn to recognise early signs of a fault. This approach can be effective, but it is time-consing to develop and often out of reach for small or niche machinery manufacturers.
Anomalyse takes a different approach. Its general-purpose machine learning platform can interpret sensor data from an industrial machine it has never seen before. It quickly learns what normal behaviour looks like and identifies patterns that suggest a fault may be developing.
However, proving that a general-purpose model works is challenging. Machine-specific models can be tested against known historical faults, showing that the model would have detected them. Anomalyse’s platform learns normal behaviour directly from live machine data. This means customers must trust that it can identify emerging faults without being trained on examples of failure.
Anomalyse’s customers needed confidence that:
To address this, Anomalyse needed a credible and validated use case that showed its platform could deliver trustworthy insights straight out of the box.
The solution: a machine failure test
Through AMPI’s Innovation for Machinery (I4M) funding, Anomalyse tested its platform on the National Physical Laboratory (NPL)’s in-house CNC milling machine. CNC milling machines are widely used in manufacturing and represent the type of industrial asset the platform is designed to support.
NPL installed two sets of vibration sensors on the machine:
The sensors were positioned side by side. This allowed a direct comparison of higher-quality and lower-quality data from the same machine, operating under the same conditions.
During the test, NPL gradually reduced the grip of the machine’s vice over successive milling runs. This simulated a realistic industrial fault that would eventually lead to a part slipping or a manufacturing process failing.
NPL colleagues measured the characteristics of the fault using a well-established vibration data analysis technique. This provided a robust reference point against which Anomalyse could compare the outputs from its own platform.
The impact: evidence to take to customers
The project showed that Anomalyse’s platform:
For example, data from the higher-quality sensors showed a clear step change in the platform’s “error score” each time the vice was loosened. This type of information could be used to estimate how close a machine is to failure, rather than simply identifying that something is wrong.
Anomalyse can now demonstrate to customers that its platform can extract meaningful signals from sensor data without explicit training, even when that data is low-grade.
This evidence base is expected to be a significant competitive advantage. It will help Anomalyse engage with customers ranging from large manufacturers with advanced monitoring systems to small and medium-sized enterprises with more basic setups. This is particularly important in a predictive maintenance market forecast to reach $60 billion globally by 2030.
What our collaborators said
Daniel Povey, Manufacturing Metrology Senior Scientist at NPL, said:
“The tests showed Anomalyse could predict the fault, even with relatively low-quality sensor data, and could offer more detailed insights with higher-quality sensors. These results will allow Anomalyse to have more nuanced conversations with customers about the level of data quality needed for the insight they want. This will help them maximise the benefit of predictive analysis tools such as those from Anomalyse.”
James Rynn, CEO and Co-founder of Anomalyse, said:
“The key value for us is credibility. This is independent, academically grounded validation that what we are building produces sensible, trustworthy outputs. As a new business, we do not yet have years of customer track record, but this gives us evidence we can stand behind when speaking to customers and investors. It is still early days, but I am confident this work will be important for future commercial opportunities.”
Supporting manufacturers across the UK
For machinery manufacturers in Greater Manchester and across the UK, Anomalyse is working to make predictive maintenance more accessible. The platform reduces the need for expensive instrumentation while increasing trust and reliability.
By lowering barriers to adoption, this approach can help manufacturers of all sizes reduce waste and downtime caused by machine drift and faults.
