Study On Approach And Real-time Implementation Technology Of On-line Surface Defect Inspection For Hot-rolled Flat Steel

With the increasing development of manufacturing industry such as modern architecture and vehicle, the producing techniques of flat steel have been improving steadily with scale. Hot-rolled strip is one of the most important products in the steel industry, and a great number of brand-new technologies have been creating in its rolling process. To increase the production efficiency while ensure the product quality level simultaneously plays a significant role on the China’s modernization process in iron and steel industry. During the five years of 2006-2010, many institutions, universities and enterprises were devoting themselves to study on the on-line surface defect inspection equipments for steel strips and had made a series of technological achievements. However, some technical issues are still besetting us especially in the rolling process and quality control of hot-rolled flat steel, the most urgent one is that the real-time performance of the current defect detection equipments could not satisfy the fast rhythm of the hot-rolling production line. As a result, the steel surfaces more or less are suffering from various defects such as roll marks, scratching or slag, and all of these defects are closely related to the raw materials and equipment conditions in upstream processes, these quality problems may lead to serious economic and reputation losses to both steel manufacturers and their customers. The main subject task about the on-line surface defect inspection for hot-rolled strip is how to accurately identify and locate the surface defects in real time through proposing what kinds of state-of-the-art defect recognition algorithms. Taking full advantages of the current technological innovations and theoretical achievements on programmable logic devices, signal processing and pattern identification, in-depth study of several related approaches and their real-time implementation technologies are discussed in this paper, which has overcome the bulky and costly shortcomings resulted from the server-array-based architecture to a large extent. This research could provide significant theoretical significance and great engineering value, not only to the steel industry but also to other metal manufacturing industries with similar high quality requirements. The main research contents and achievements of this dissertation are as follows:1. In order to develop a kind of on-line surface defect inspection equipment with high reliability and wide adaptability, we had spent a great deal of time and effort on the field investigation and technical exchange with lots of hot continuous rolling mills at the early stage of this research, and a cost-effective system scheme was proposed after that the overall design goals which could cover the average requirements of global hot-rolling mills had been established. In accordance with the demands of industrial products, the whole system integration has been finished in this paper so that it could be adapted to the awful environment on the production line suffered with high temperature, humidity and multi dust. Aiming at the huge image data stream(its peak rate is high up to 4.0 Gbps), a 12-layer customized image processing board based on FPGA was developed, which provids a design idea and hardware scheme for real-time image processing of the strip surface image, this implementation approach can be competent for production line with rolling speed high up to 20 m/s. The key techniques about PCB layout, driver program and embedded software were also discussed in detail.2. This paper has arranged considerable length on image enhancement algorithm. A series of contrastive experiments were carried out for studying the enhancement effects of popular methods including histogram equalization, homomorphic filtering, Retinex and gradient domain algorithm. A novel dynamic homogeneous compensation(DHC) algorithm based on median filtering was proposed, and the experimental results prove that this algorithm has effectively handled the uneven illumination while maintaining defect details well, whether surface images are suffering from uneven-lighting, over- or under-exposure. In addition, it is a kind of completely self-learning algorithm and with area-saving advantage.3. Based on the variance, entropy and average gradient of the 32×32 pixel image block, we proposed an adaptive dual thresholding(ADT) algorithm to identify the surface defect in real time, the variance and average gradient threshold parameters were dynamically amended in accordance with the sensitive entropy. The ADT algorithm belongs to threshold discrimination method and can be implemented on FPGA easily, which accomplishes the fast defect extraction on the underlying data layer.4. FPGA-based hardware implementation architecture for image cache, image enhancement and defect recognition algorithms has been developed in this paper. A kind of DMA transaction queue management strategy based on ring FIFO was proposed and applied on the large-scale image data transmission and PCIe interface management. Moreover, we designed a time scope(TS) tool to achieve quantitative analysis about the real-time performance of all the embedded functional modules.5. The final designed on-line surface defect inspection prototype for hot-rolled strip has been tested and upgraded on the hot-rolling production line in Valin LY Steel more than 3 years. The application performance illustrated that the achieved average true positive rate(TPR), false positive rate(FPR), false negative rate(FNR) and location precision(LP) have satisfied the expected performance goals, while the integrated cost is only around 1/6 of that of imported equipment possessing similar performance, which provides theoretical basis and a case study for the surface quality control of flat steel with our own intellectual property rights in China.