CN-122016873-A - Millimeter wave radar boiler heating surface pipe crack detection method based on GPU parallelization

Abstract

The method comprises the steps of constructing a three-layer parallelization frame integrating data flow, calculation and control, and utilizing the massive parallel computing capacity of the GPU to perform full-flow acceleration processing from preprocessing, range profile reconstruction and feature extraction to intelligent recognition on massive echo signals acquired by the millimeter wave radar, so that high-precision and high-real-time online detection and positioning of micro cracks on the wall of a heating surface are realized in a complex environment in a strong-interference boiler, the technical effects of improving the detection efficiency and simultaneously keeping high detection confidence are finally achieved, and powerful technical support is provided for safe operation and preventive maintenance of a power plant boiler.

Inventors

• HOU ZHAOTANG
• HU SHIMING
• LI XIN
• WANG ZONGYU
• GAO LEI
• QIAN SHIYOU
• LIAO ZHENGYU
• CHEN YICHAO
• LI JIAOJIAO
• ZHANG FUXIANG
• WANG PENG
• ZHANG HAO
• CAI ZUXIN
• CAO ZHIHUA
• ZHANG ZENGHUI
• Lv ziyang

Assignees

• 西安热工研究院有限公司
• 华能(福建)能源开发有限公司福州分公司

Dates

Publication Date

20260512

Application Date

20260410

Claims (6)

1. 1. The millimeter wave radar boiler heating surface pipe crack detection method based on GPU parallelization is characterized by comprising the following steps of: S1, scanning the wall of a heating surface of a boiler through a millimeter wave radar, collecting original echo signal data, and transmitting the original echo signal data to a GPU video memory; S2, on a GPU, performing signal preprocessing on the original echo signal data through parallel calculation, generating a range profile, performing parallelized feature extraction on the range profile, obtaining a crack candidate region, and performing parallel recognition on the crack candidate region by using a pre-trained neural network model to obtain a crack detection result; And S3, outputting and visualizing the crack detection result.
2. 2. The method of claim 1, wherein the signal preprocessing of the raw echo signal data by parallel computing comprises: Performing parallelized amplitude normalization and denoising treatment on the original echo signal data by utilizing a CUDA thread block; Window function weighting is carried out on the denoised signals by utilizing the CUDA kernel function and the shared memory; invoking a batch FFT plan of cuFFT library, and performing fast Fourier transform on the weighted signals in parallel to generate a range profile; and calculating the power spectrum of the range profile in parallel.
3. 3. The method of claim 1, wherein the feature extraction that parallelizes the range profile comprises: calculating the neighborhood direction gradient and gradient amplitude of each pixel of the range profile through parallel calculation; Dynamically calculating a threshold value based on the local mean value and the standard deviation, and judging in parallel and generating a mask matrix of crack candidate pixels; Adopting a connected domain marking algorithm based on parallel union, carrying out parallel connected domain analysis on the crack candidate pixels, and identifying a crack region; and performing parallel post-processing on the identified crack region through the GPU morphological filtering kernel function.
4. 4. The method of claim 1, wherein the concurrently identifying the crack candidate region with a pre-trained neural network model comprises: inputting the crack candidate region into a TensorRT optimized lightweight CNN model; on the GPU, forward reasoning of the model is executed in parallel by utilizing multiple CUDA streams and a batch input mode; and outputting confidence, position and size information of the crack.
5. 5. An electronic device, comprising: one or more processors; a storage unit configured to store one or more programs, which when executed by the one or more processors, enable the one or more processors to implement the GPU-parallelized-based millimeter wave radar boiler heating surface pipe crack detection method according to any one of claims 1 to 4.
6. 6. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, is capable of implementing the GPU-parallelized-based millimeter wave radar boiler heating surface pipe crack detection method according to any one of claims 1 to 4.

Description

Millimeter wave radar boiler heating surface pipe crack detection method based on GPU parallelization Technical Field The invention relates to the technical field of industrial defect detection, in particular to a millimeter wave radar boiler heating surface pipe crack detection method based on GPU parallelization. Background The boiler is a key device of a thermal energy system of a power plant, and a heating surface pipe of the boiler is in a high-temperature and high-pressure environment for a long time, so that structural defects such as microcracks and the like are extremely easy to occur. If the cracks are not found and treated in time, boiler leakage and even explosion accidents can be caused. The traditional crack detection method mainly comprises ultrasonic detection, infrared thermal image detection and manual inspection. These methods suffer from the following drawbacks: the detection precision is limited, namely the infrared thermal imaging method is greatly influenced by temperature distribution, and the identification of micro cracks is difficult under a strong interference environment; The adaptability is poor, the ultrasonic detection needs to be contacted with the wall surface, and the ultrasonic detection is difficult to be applied on line in real time in a high-temperature environment; The real-time performance is insufficient, and the manual inspection and serial signal processing modes can not meet the requirement of large-scale rapid detection during the operation of the boiler. Millimeter wave radars have become a potential scheme for boiler detection due to the advantages of penetrating smoke, non-contact detection, sub-millimeter resolution and the like. However, in the millimeter wave echo signal processing process, the data volume is huge, and the real-time detection is difficult to realize by the traditional CPU serial processing mode. The existing method lacks a GPU parallelization frame design and crack recognition algorithm optimization scheme aiming at radar data. Disclosure of Invention In a first aspect of the present disclosure, a method for detecting cracks on a heating surface tube of a millimeter wave radar boiler based on GPU parallelization is provided, including the following steps: S1, scanning the wall of a heating surface of a boiler through a millimeter wave radar, collecting original echo signal data, and transmitting the original echo signal data to a GPU video memory; S2, on a GPU, performing signal preprocessing on the original echo signal data through parallel calculation, generating a range profile, performing parallelized feature extraction on the range profile, obtaining a crack candidate region, and performing parallel recognition on the crack candidate region by using a pre-trained neural network model to obtain a crack detection result; And S3, outputting and visualizing the crack detection result. With reference to the first aspect, the performing signal preprocessing on the original echo signal data through parallel computation includes: Performing parallelized amplitude normalization and denoising treatment on the original echo signal data by utilizing a CUDA thread block; Window function weighting is carried out on the denoised signals by utilizing the CUDA kernel function and the shared memory; invoking a batch FFT plan of cuFFT library, and performing fast Fourier transform on the weighted signals in parallel to generate a range profile; and calculating the power spectrum of the range profile in parallel. With reference to the first aspect, the feature extraction for parallelizing the range profile includes: calculating the neighborhood direction gradient and gradient amplitude of each pixel of the range profile through parallel calculation; Dynamically calculating a threshold value based on the local mean value and the standard deviation, and judging in parallel and generating a mask matrix of crack candidate pixels; Adopting a connected domain marking algorithm based on parallel union, carrying out parallel connected domain analysis on the crack candidate pixels, and identifying a crack region; and performing parallel post-processing on the identified crack region through the GPU morphological filtering kernel function. With reference to the first aspect, the identifying the crack candidate region in parallel by using a pre-trained neural network model includes: inputting the crack candidate region into a TensorRT optimized lightweight CNN model; on the GPU, forward reasoning of the model is executed in parallel by utilizing multiple CUDA streams and a batch input mode; and outputting confidence, position and size information of the crack. In a second aspect of the present disclosure, a millimeter wave radar boiler heating surface pipe crack detection system based on GPU parallelization is provided, including: The millimeter wave radar signal acquisition module is used for controlling the radar probe array to scan the wall of the heating surfac