Search

• Diagnosis of twenty faults in refrigeration systems by one classifier is proposed. • CNN, LDA, SVM, PCA-SVM, LDA-SVM classifiers are compared. • Variation of training data is very important to achieve better classification result. • CNN and PCA-SVM can not diagnose the faults satisfactorily. • SVM obtained the best verification result. The functionality of industrial refrigeration systems is important for environment-friendly companies and organizations, since faulty systems can impact human health by lowering food quality, cause pollution, and even lead to increased global warming. Therefore, in this industry, there is a high demand among manufacturers for early and automatic fault diagnosis. In this paper, different machine learning classifiers are tested to find the best solution for diagnosing twenty faults possibly encountered in such systems. All sensor faults and some relevant component faults are simulated in a high fidelity Matlab/Simscape model of the system, which has previously been used for controller development and verification. In this work, Convolutional Neural Networks, Support Vector Machines (SVM), Principal Components Analysis-SVM, Linear Discriminant Analysis-SVM, and Linear Discriminant Analysis classifiers are compared. The results indicate that the fault detection reliability of the algorithms highly depends on how well the training data covers the operation regime. Furthermore, it is found that a well-trained SVM can simultaneously classify twenty types of fault with 95% accuracy when the verification data is taken from different system configurations. [ABSTRACT FROM AUTHOR]

Abstract: The primary goal was to evaluate and enhance a virtual refrigerant charge (VRC) sensor developed by Li and Braun (2006, 2009). Based on the evaluation, the VRC sensor was found to work well in estimating refrigerant charge for systems that do not utilize accumulators when using the default parameters. For systems with accumulators, however, the parameters needed to be improved for better accuracy. To overcome those limitations, a simulation method for estimating parameters was developed that depends on test data at rated conditions and system information. The simulation parameters decreased the errors between the actual and predicted charge. Even better performance was achieved for the VRC sensor when the parameters were tuned by minimizing charge prediction errors using test data and linear regression. The overall RMS error for the VRC sensors using tuned parameters was 3.8% over a wide range of operating conditions for different types of systems. [Copyright &y& Elsevier]

Abstract: The benefits of applying automated fault detection and diagnosis (AFDD) to chillers include less expensive repairs, timely maintenance, and shorter downtimes. This study employs feature selection (FS) techniques, such as mutual-information-based filter and genetic-algorithm-based wrapper, to help search for the important sensors in data driven chiller FDD applications, so as to improve FDD performance while saving initial sensor cost. The ‘one-against-one’ multi-class support vector machine (SVM) is adopted as a FDD tool. The results show that the eight features/sensors, centered around the core refrigeration cycle and selected by the GA-SVM wrapper from the original 64 features, outperform the other three feature subsets by the GA-LDA (linear discriminant analysis) wrapper, with an overall classification correct rate (CR) as high as 99.53% for the 4000 test samples randomly covering the normal and seven typical faulty modes. The CRs for the four cases with FS are all higher than that without FS (97.45%) and the test time is much less, about 28–36%. The FDD performance for normal or each of the faulty modes is also evaluated in details in terms of hit rate (HR) and false alarm rate (FAR). [Copyright &y& Elsevier]

Abstract: We report development of a micro thermal sensor and sensing techniques for monitoring the oil concentration in a refrigeration system. The sensor acquires the thermal response of the mixture to ac heating and then the concentration can be determined using one of the two different methods. In the first method, the thermal conductivity of the mixture is obtained using the three-omega method and the concentration is determined using the correlation between the mixture concentration and thermal conductivity. In the alternative fast-detection method, the concentration is determined by directly calibrating the sensor output signal to the mixture concentration. The performance of the sensor was tested using the R410A/polyvinyl ether (PVE) oil mixture. The uncertainty of the oil concentration measurement was estimated to be 5.8–7.3 wt.%, depending on the sensing scheme. Especially, the simple structure of the sensor makes the technique cost effective and adequate for miniaturization, i.e., for installation in a limited space. [ABSTRACT FROM AUTHOR]

Abstract: Both vertically and horizontally installed heat source systems for ground-coupled heat pumps based on the thermosyphon principle using CO2 as the working fluid have been investigated theoretically as well as experimentally. The investigation of the vertical type (CO2 probe) has started in 1997. During the last years various design and operational parameters have been analyzed in depth, e.g. optimum number of parallel earth probes, probe length, CO2 charge required etc. In Austria up to now more than 100 systems have been installed, and the experience confirms that a heat pump system consisting of heat pump, probe head, and CO2 probe is a reliable, highly efficient as well as environmentally friendly alternative to commonly used ground-coupled systems. The great success with the CO2 probe initiated further research activities on a horizontal type (CO2 collector). The first experimental results carried out with a test rig in the laboratory confirm that CO2 can be used as heat carrier in collectors. [Copyright &y& Elsevier]

It is first recalled that the temperature profile in a freezing product is almost linear. To estimate the freezing level of a food product, a non-intrusive sensor is proposed which may estimate the surface temperature and the slope of the internal temperature profile. It consists in analysing transient response when the sensor is set in contact with the product to characterise. The sensor is made of a thermopile (Peltier element) fixed on an isothermal medium made of a high effusivity material (copper). A modelling of the transient response of the sensor using the quadrupoles method is presented. Then, the results of the characterisation of the sensor by experimental determination of the transfer functions between fluxes and temperatures are given. A theoretical sensitivity analysis enables to study of the potential limits of this sensor. Finally, the influence of internal temperature profile slope on the sensor response is evidenced by the results of an experimental study. The main interests of this sensor are its simplicity and low cost and a flux measurement independent of thermal contact resistance to estimate a temperature. [Copyright &y& Elsevier]