Heat pumps represent an effective technology to convert electrical energy to thermal energy through the expansion and compression of a refrigerant. Heat pumps are becoming increasingly more efficient which decreases energy consumption. However as heat pumps are getting increasingly more efficient, complexity is also increasing which leads to failures.

GEA is one of Europe’s largest providers of industrial heat pumps, delivering a variety of different industrial heat pumps for food production, breweries, waste heat recovery, and thermal power plants. As industrial heat pump systems are becoming increasingly complex, so is the task of detecting faults in the systems. Whenever a heat pump breaks down it is challenging to see what the reason for the failure is without visual inspection. A typical service engineer will have data from over 20 sensors, each with over two and a half million measurement intervals, to inspect and look for the cause of failure. Visual inspection of the heat pump systems can only be done by experienced engineers who knows
exactly what to look for and where to look for it.

As heat pumps have been developing so has the sensor technology and computational resources available in modern computers. Over the last decades the price of sensors has decreased rapidly while the capability and reliability of sensors has improved. The computational power of modern computers has increased significantly while the price for these resources has decreased. 
Several other industries have, with the development in sensor technology and computational power, investigated the possibilities of implementing Fault Detection & Diagnosis (FDD) programs to automatically
analyze sensor data in order to detect failures and diagnose the failure cause. FDD programs were previously limited to critical processes, for example power plants, aerospace, or national defence. They
were expensive programs due to the sensor requirements, the computational resource requirements and the development costs. However, as the price of sensors and computational resources has been
decreasing, FDD is becoming increasingly more popular for less critical equipment. For the past two decades FDD programs have been developed for different Heating, Ventilation, Air-Conditioning and Refrigeration (HVAC&R) equipment with a special focus to refrigeration.

This project aims to develop a FDD program for GEAs industrial heat pump systems. Following a literature study it was concluded that thus far, FDD programs have not been utilized for heat pump systems of this complexity. It was seen that most FDD programs use a quantitative, qualitative or process history based approach to detect faults. Quantitative FDD programs aim to design a digital twin of the system at hand based on physical relations, and compare measured values to defined values for
faulty or fault-free operation. Qualitative FDD programs aim to mimick the behaviour of system experts and base decisions on previously seen failures and cases. Process history based FDD programs aim
to learn how systems should ideally behave through an abundance of operational data, and compare systems to sampled data.

The FDD program for GEA uses a qualitative approaches for all faults where expert knowledge is available. For faults not seen before, a process history based approach classifies operation as faulty or
fault-free. The complexity of the faults and the system do not allow for a quantitative FDD program. A predefined list of in total 23 faults was agreed upon, the FDD program further aims to detect all 23 failures and diagnose the failure cause.
The project is meant as a proof of concept for GEA. By proving that an off-line FDD program is possible, GEA can ultimately implement the algorithms used in this off-line FDD program in to the control system of the heat pumps to increase the reliability of their heat pumps and detect faults as they are occurring.