What if a heating system could be monitored not only through conventional sensors, but also through patterns visible in thermal images?
That question is at the heart of JOULIA’s Deliverable 7.2, “Predictive maintenance platform”, led by CERTH with contributions from ART, SIML, STAM, K-FLEX and KTH. The work describes the project’s initial research towards a platform that can monitor heating-system behaviour using infrared thermography and, over time, support predictions and alerts for abnormal conditions.
The approach starts with a practical challenge: industrial thermal images are not automatically clean data. Reflections, surface properties and changing operating conditions can make apparent temperatures difficult to interpret. D7.2 therefore examines image pre-processing and processing techniques designed to improve the usefulness of thermal data before anomaly detection is attempted.
The project has also collected baseline thermal imagery in the industrial environments that will later host JOULIA’s electrified heating technologies. These observations help the team understand where thermography can be useful, what visual artefacts need to be handled and what additional data will be needed as the new systems are introduced.
Importantly, D7.2 presents a preliminary approach rather than a finished predictive-maintenance product. Further image collection, processing and comparison will continue as the project advances.
For industrial stakeholders, the direction is compelling: electrification and digitalisation can progress together, with monitoring tools helping future operators detect unusual thermal behaviour earlier and build confidence in new heating technologies.
What the public deliverable actually tests: D7.2 is the first stage of the predictive-maintenance research, not a claim that a finished platform has already been validated. CERTH and the contributing partners first examined how infrared thermography could be applied to the current production environments at ARTIGO and K-FLEX. Thermal acquisitions were used to understand what can be observed reliably, where measurement conditions are challenging and what image-processing steps may be needed before anomaly detection can be trusted.
The proposed platform concept uses thermal images from relevant parts of the production lines as a reference or ground truth. Once the new JOULIA heating systems are implemented, future images can be compared with that reference to look for abnormal thermal behaviour. The public deliverable describes a preliminary two-stage route: image pre-processing to improve the quality of the thermal data, followed by anomaly detection using image-processing and machine-learning techniques.
For industrial stakeholders, the significance is practical. Predictive maintenance is most useful when it is built on data that actually reflect the conditions of the plant. D7.2 therefore starts with measurement reality: camera placement, surface behaviour, reflections and access, before moving towards algorithms. That sequence can improve the credibility of future alerts and reduce the risk of confusing measurement artefacts with genuine process or equipment problems.
Explore JOULIA’s D7.2 and follow the next phase of predictive-maintenance development.
