Portuguese researchers have compared different machine learning techniques applied to detect faults in air-to-air heat pumps in cooling mode. The results showed a high level of performance based on four measures.
A Portuguese research team has compared different applied machine learning techniques to perform fault detection in air-to-air heat pumps in cooling mode and has identified four algorithms that show a high level of performance based on the metrics accuracy, precision, recall and F1 score, a key figure used to measure classifier performance.
According to the research team, the strong growth in the use of heat pumps in heating, ventilation and air conditioning (HVAC) systems of buildings creates a demand for ease of maintenance and reliability through early detection of system errors.
“The use of artificial intelligence (AI) tools is gradually evolving, making it able to potentialize development in different areas,” said the corresponding author of the study, Pedro Barandier. pv magazineexplaining that the research was motivated by the “remarkable growth” of heat pumps and the promise of classification algorithms, neural networks and other AI tools to improve the reliability of such systems, especially when used alongside Internet of Things technologies used.
Python was used to perform the comparative analysis of supervised learning classification algorithms for fault detection. The algorithms studied include Naïve Bayes, Support Vector Machine, Logistic Regression and K-nearest neighbors with the plan to evaluate them based on accuracy, precision, recall and F1 score. a measure of the harmonic mean of precision and recall.
The team chose typical cooling mode faults such as compressor and reversing valve leakage, improper contamination of the condenser and evaporator, liquid line blockage, too little or too much refrigerant, and the presence of non-condensable gases.
A data set derived from previous research using an 8.8 kW (HP) residential heat pump, with a scroll compressor and a thermostatic expansion device, was used. It contained 96 variables and was made up of 7,374 tests performed in cooling mode, including temperatures, pressures, air and coolant mass flow rates, electrical power, coefficients of performance, error levels and more. Several parameters of the heat pump’s thermodynamic circuit were also present, such as temperatures, pressures, mass flow rate and electrical power of the compressor.
Several rounds of feature selection were conducted. In addition, a principal components analysis was also conducted, and it was determined that ten components explained more than 90% of the variance. To identify the variables with the most impact on the results, an ablation study was also performed, as well as a correlation matrix to further reduce the number of features.
A connection weight approach was performed through a single-layer artificial neural network and 40 features were selected. Moreover, a synthetic minority oversampling technique (SMOTE) was performed to balance the healthy cases in the training set.
The best results were achieved with K-nearest neighbors, which showed the highest values for each of the four metrics, above 99% after a cross-validation considering a smaller number of components. But Naïve Bayes, Support Vector Machine and Logistic Regression also achieved performance metrics above 90%.
Other models were superficially analyzed using the machine learning tool PyCaret, according to the team. “Ridge Regression is on average the fastest of the best performing algorithms with high values for accuracy, precision, recall and F1. This is an important finding considering that no previous applications of this algorithm for the diagnosis of HP faults have been found in the literature,” the report concluded.
Looking ahead, the research team will go beyond just error detection. “Because the fault diagnosis process depends on several intrinsic aspects, this work only looked at the first one: fault detection. The team’s next goal is to further develop the knowledge acquired to develop a complete and efficient fault diagnosis system for heat pumps that can be exploited by engineers and professionals around the world,” said Barandier.
The research is described in detail in “Comparative analysis of four classification algorithms for heat pump fault detection,” published by Energy and Buildings. The researchers came from the Portuguese University of Beira Interior and the Polytechnic University of Coimbra.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact: editors@pv-magazine.com.