Performance improvement in CO2 heat pump dryers

Greenhouse gas effects, global warming and ozone layer depletion which effects whole world and caused by harmful refrigerants, are very important and dangerous facts. There are many regulations and incentives are being prepared, in order to slow down and decrease that danger. Companies and universities rapidly started trying to use environmentalist yet more natural refrigerants in heat pumps and air conditioners. Among all those refrigerants, hence it is natural and as well as its thermophysical properties, CO2 is one of the most preferred refrigerant. In this study, transcritical cycle based CO2 (R744) heat pump dryer performance, electric consumption and drying duration can be calculated via neural network model. Previously created mathematical MATLAB model's inputs and outputs were used for newly created neural network model to learn. When model was learnt the system once, it can easily response unknown or unidentified inputs. Two main feedforward neural network model were created. First one is the model that is learning the relations between mathematical MATLAB model's inputs and outputs, and testing the results; the other model is the test model. System was taught to the test model via 33 cases extracted mathematical model. The extracted 33 cases' inputs were tested on the narrowed space (33 case extracted) model and results were compared. All neural networks models learn the inputs and outputs with a transfer function. 3 different transfer functions (linear, sigmoid and tangent hyperbolic) were tried both xvii input-output couple. Best transfer function couple was chosen to optimize for each model. The maximum of average errors of the 12 outputs were found 10.795% in comparison of non-optimized first main models and mathematical model. At the optimized main model side; the maximum of average errors of the 12 outputs were found 5.40%. At the optimized test model comparison, the maximum of average errors was found 21.498% -as expected due to narrowed space. With this work, it was obtained that; by usage of neural network codes that embedded in dryers, can improve consumption of energy and drying duration in real time.