Skip to main content

Case Study

Mitsubishi + Glasgow Caledonian University

Development and implementation of a model for monitoring and optimisation of production processes.                

Mitsubishi

Mitsubishi Electric Air Conditioning Systems Europe Ltd (M-ACE) manufactures air-conditioning and heat-pump systems as part of the global Mitsubishi Electric (ME) portfolio of products, supplying mainly Europe, Middle East and Africa. 
 

What was the need?

The Challenge

The project faced several challenges, beginning with the limitations of the aging equipment. The Power Press motors, being around 30 years old, initially lacked the necessary sensors for capturing vibration or temperature data. Installing new Bluetooth vibration sensors, integrating them with the existing IoT infrastructure, and establishing a reliable data collection process required significant setup and calibration efforts.

Data preprocessing also posed challenges, particularly in distinguishing between genuine anomalies and normal vibrations caused by the press’s metalworking operations. High vibration levels during metal pressing were common and could be misinterpreted as faults, so the team had to develop filtering techniques to focus only on motor health indicators. Additionally, tuning and optimizing the machine learning models required careful consideration, especially given the complexity of the data and the need to select the most suitable algorithms. Finally, setting actionable thresholds for predictive alerts involved balancing sensitivity with the practical need to avoid false alarms, ensuring maintenance interventions were timely yet efficient.

What did we do?

The Solution

The solution involved implementing a predictive maintenance model that transitioned the maintenance approach from reactive to proactive. Bluetooth-enabled vibration sensors were installed on the motor, allowing for real-time monitoring of operational health. These sensors connected to the facility’s IoT infrastructure via gateways, which transmitted data to a centralized Cassandra database for storage and processing.

To accurately detect potential motor issues, data preprocessing techniques were applied to filter out noise caused by external operations, such as the press’s metalworking activities. The model then analyzed six months of historical vibration data using multiple algorithms—ARIMA, Random Forest, and LSTM—with ARIMA selected as the most effective for predicting motor health. By setting an anomaly threshold based on residual analysis, the model could alert the maintenance team when vibration patterns indicated an impending issue, allowing them to address it before failure occurred.

What changed?

The Impacts and Benefits

Impacts for the Company 

The predictive maintenance model can improve M-ACE’s efficiency and lower costs by reducing unplanned downtime and enabling timely, data-driven repairs. Real-time motor monitoring will allow early issue detection, extending equipment life and boosting reliability. This proactive approach also can enhance productivity and set the foundation for expanding predictive maintenance across other factory assets.

 

Impacts for the Academic Team

The project’s presentation at IMEKO 2024 and submission to Measurement: Sensors journal enhanced the academic team’s visibility and impact, showcasing their expertise in predictive maintenance and advancing research in industrial IoT applications.

 

Impacts for the KTP Associate

Completing this project deepened my expertise in predictive maintenance and IoT, especially in sensor installation and data collection processes for industrial equipment. Working hands-on with vibration sensors, data preprocessing, and machine learning models enhanced my skills in data-driven maintenance strategies. Presenting the work at IMEKO 2024 also boosted my visibility in the field and added to my professional credibility as a specialist in applying advanced analytics to improve manufacturing efficiency.

The Impacts and Benefits

The People

Meet the Team

Dr Shahram Hanifi

KTP Associate

LinkedIn

Professor Babakalli Alkali

Knowledge Base Supervisor

LinkedIn

Allan Borthwick

Company Supervisor

LinkedIn