Applications of artificial intelligence to refrigeration plants

At the IIR Conference on Thermophysical properties and Transfer Processes of Refrigerants held in September 2021 in Vicenza, Italian researchers presented a performance review of 19 refrigerating plants in which Artificial Intelligence algorithms have been implemented. (1) 

 

Artificial Intelligence (AI), including Machine Learning (ML), Artificial Neural Networks (ANN) and Deep Learning (DL) – which are subfields of AI – can bring significant benefits in terms of the operation and maintenance of refrigeration and air-conditioning systems. 

 

Fault detection and diagnosis (FDD) for predictive maintenance and improvement of machine operations is one of the most studied AI  applications in these fields. As predictable faults, the authors give the examples of refrigerant charge defects, in terms of overcharge or leakage/undercharge, and heat exchangers failures in terms of fouling, secondary fluid mass flow and heat transfer reduction. The use of AI algorithms instead of physics-based ones can be interesting if reduced computation time is required and if operational cumulative effects are considered in time. FDD allows for better scheduling of maintenance operations in order to avoid serious failures and to limit the costs related to minor failures. 

 

AI is also becoming widely used for energy performance prediction of refrigeration equipment. Energy performance prediction is useful for establishing the optimal control strategy and maximising energy savings. For example, predicting the COP of a ground-source heat pump with an ANN algorithm using ground temperature, inlet and outlet condenser air temperature as input variables resulted in an error of only 1% between the measured and predicted COP. 

 

Predictive control is an important application of AI. Thanks to their simplicity and reduced computation time, ANN algorithms can be implemented in the main control boards of systems in real time operation. These novel control systems can overcome the limits of the widely used thermostatic and PID controlsby providing shorter response time during transient operation, achieving significant energy savings. As an example, the authors present the modeling of a 4-chiller plant with a two-level algorithm to minimise operating costs. At the first level, a genetic algorithm is used to predict the on/off status of the chiller based on the cooling load. At the second level, particle swarm optimisation is used to minimise the power consumption of the system using cooling capacity, water temperature difference and enthalpy as the main inputs. Over two days of operation, an energy saving of 14% was achieved; moreover, thanks to a high calculation rate, it is possible implement this algorithm to control the entire plant in real time.  

 

On the other hand, little work has been done on the prediction of frost formation and on defrostng control optimisation techniques, showing that this area of research remains to be explored.

 

All papers and proceedings from the IIR 2021 TPTPR conferences can be downloaded using the following links:

 

(1) Citarella B. et al, Use of Artificial Intelligence in the refrigeration field, IIR 2021 TPTR conference: link.