A machine learning bayesian network for refrigerant charge faults of variable refrigerant flow air conditioning system

Review on Fault Detection and Diagnosis Feature Engineering in Building Heating, Ventilation, Air Conditioning and Refrigeration Systems

“…Filter selects FFs from the entire original variable set alone prior to the FDD modeling process. In terms of the HVAC&Rs, it attempts to evaluate system data variables and their sensitivities to fault operations using various measures including mutual information [72,221,250], Pearson correlation-based coefficient [86,88,89,236,237,249,250,252], inter-class and intra-class distances [72], Relief and ReliefF indexes [226,227,242,251], information gain and the gain ratio [46] and the rough set algorithm [239]. Variables that are evaluated with high-scored measures should be selected as FFs using pre-defined thresholds.…”

“…The parallel combination performs FS algorithms simultaneously and determines the FFs that all FSs select. Hu et al [46] combined five filters and most frequently selected variables were adopted for VRF FDD modeling. However, for the twostep serial combination method, the filter is performed first, prior to integration with the wrapper or embedded technique.…”

“…Recently, it has become clear that -compared with the quantitative and qualitative model-based methods, -the data-based FDD methods are superior especially for largescale HVAC&Rs [29]. For applications in HVAC&Rs, FDD studies have implemented several data analytics algorithms, including machine learning [30][31][32][33], data mining [34][35][36][37], statistics [38][39][40], decision-making [41][42][43][44][45][46], expert system [40,47,48] and pattern recognition [49,50] algorithms. From the perspective of data analytics, the objective of fault detection is to ascertain whether a given data sample lies Feature engineering (FE) is an important step in generating the optimal model inputs for the FDD strategy in HVAC&Rs.…”

Rapidly Learning Bayesian Networks for Complex System Diagnosis: A Reinforcement Learning Directed Greedy Search Approach

“…For the development of sensor and data storage techniques, data-driven diagnosis (DDD) [1]- [6] received much attention recently. As one of the most popular data-driven methods, Bayesian networks (BNs) are widely applied to fault diagnosis [7]- [12].…”

“…Their BNs was composed of three layers: causal factors in layer 1, faults in layer 2 and fault symptoms in layer 3. In the diagnosis of refrigerant flow air conditioning systems, Hu et al [12] developed a three-layer BN model as well: faults in the first layer, features in the second layer, and additional information in the third layer. In the two researches, posterior probabilities of faults under measurements were calculated for fault diagnosis.…”