| CPC F24F 11/38 (2018.01) [F24F 11/46 (2018.01); F24F 11/64 (2018.01); G06N 3/0464 (2023.01); G06N 3/082 (2013.01)] | 8 Claims |
|
1. A fault detection method for refrigeration units based on an improved deep learning model, comprising the following steps:
providing a refrigeration unit: wherein the refrigeration unit is a heat-pump air conditioning apparatus: the heat-pump air conditioning apparatus comprises an evaporator, a condenser, and a thermal expansion valve: wherein the evaporator and the condenser comprises air-cooled finned-tube heat exchangers: the thermal expansion valve serves as the throttling device; and a thermal-sensing bulb for the expansion valve is mounted at the compressor inlet:
S1: obtaining operating parameters of a refrigeration unit in a normal operating state and in states with different fault types as data sets, by a processor;
S2: detecting local outliers in the data set by using a local outlier factor algorithm and removing the local outliers, and then expanding the data set by using adaptive synthetic sampling, by a processor;
S3: normalizing the data set, by a processor;
S4: constructing a fault detection model, by a processor, wherein the fault detection model comprises a ResNet module and a CBAM module, the ResNet module comprises three residual blocks connected in sequence, each residual block is introduced with a Dropout layer, the Dropout layer randomly closes some neurons of a 1DCNN layer connected to it at a dropout rate p, and input of the residual block and output of the Dropout layer are identity mapped and added together as output of the residual block;
output of the ResNet module serves as input of the CBAM module; the CBAM module includes sequentially connected channel attention module and spatial attention module; in the spatial attention module, a channel attention weight output by the channel attention module are subjected to max pooling and average pooling on channels of each feature point to obtain feature maps Fmaxc and Favgc, respectively; and then the feature maps Fmaxc and Favgc are concatenated based on the channels, and obtained features are subjected to convolution operations with three different scales; three results of the convolution operations are fuse and activated; sizes n1, n2 and n3 of convolution kernels of the convolution operations with three different scales satisfy a following constraint: n1+2=n2=n3−2, and n1, n2 and n3 are positive integers;
an output feature of the CBAM module is input to fully connected layers, and the fault detection model is trained by using the data set obtain in the S2; and
S5: inputting the parameters of a to-be-tested refrigeration unit into the fault detection model, and generating a table layout that allows visualization of the fault detection result.
|