With the development of the new energy vehicle industry, laser welding has been widely used in the manufacturing of power batteries for its fast welding speed, small heat-affected zones, and high degree of automation. During laser welding, the welding depth fluctuates because of the involved process parameters and impact factors, which produce defects such as insufficient welding depth and burn through. In order to avoid these defects, online monitoring of the laser welding depth is necessary to achieve quality monitoring. The indirect methods based on the various signals such as optical radiation, visual image, and acoustic wave in the process zone are easily affected by the welding process and cannot ensure high accuracy of the welding depth measurement. A laser welding depth monitoring method based on optical coherence tomography (OCT) directly measures the depth of the keyhole in the molten pool under deep-penetration welding conditions. This method is based on low coherence interferometry and aligns the measuring beam with the processing beam, which has the advantages of high measurement accuracy and strong anti-interference ability. The dynamic changes in the keyhole lead to the scattered distribution of the raw OCT data, and the post-processing of the raw OCT data is required to reveal the welding depth. The percentile filter is confirmed to be viable to process the raw OCT data. The accuracy of the welding depth extraction by using this method is easily affected by the noise of raw OCT data, and filter parameters are needed to adapt to different welding conditions. In order to solve the aforementioned problems, an OCT welding depth extraction method based on local outlier factor (LOF) and maximum filter is proposed in this paper. This is done by first detecting the noise points of raw OCT data using LOF and then applying the maximum filter to the cleaned OCT data.

Firstly, this paper builds a laser welding depth measurement system based on SD-OCT and measures the keyhole depth under deep-penetration welding conditions. A longitudinal cross section of the weld seam is obtained to extract the actual welding depth. Then, the percentile filter is applied to the raw OCT data to extract the welding depth. In order to evaluate the accuracy of the welding depth extraction, the average error is introduced by considering the difference between the welding depth extracted from OCT data and the actual weld depth extracted from the longitudinal cross section. Next, the proposed method is applied to the raw OCT data to extract the welding depth. The noise points of the raw OCT data are first detected by LOF and removed from the raw OCT data. The welding depth is then extracted by the maximum filter. The proposed method is compared with the percentile filter in terms of the average error. Finally, in order to verify the repeatability of the proposed method, multiple welds with the same process parameters are performed.

A laser welding depth measurement system based on SD-OCT is established, and during the welding process, the raw OCT data are obtained (Fig. 4). Firstly, the percentile filter is applied to the raw OCT data to extract the welding depth. An average error of less than 5% can be achieved with a percentile between 92 and 98 and a window length greater than 200. Next, in order to further improve the accuracy of welding depth extraction, the proposed method is applied to the raw OCT data to extract the welding depth. The noise points of the raw OCT data are first detected by LOF, and the maximum filter is then used to extract the welding depth (Fig. 12). Furthermore, the comparison in average error between the proposed method and the percentile filter is conducted. The average error is 3.0% in the proposed method and 4.4% in the percentile filter (Fig. 13). Finally, multiple welds with the same process parameters are performed, and the proposed method is applied to extract the welding depth. The average errors are 3.0%, 3.8%, and 3.4%, respectively. Compared with the percentile filter, the accuracy of the welding depth extraction is improved by 32%, 22%, and 24%, respectively (Table 1). Therefore, it can be concluded that an average error of less than 4% can be achieved by the proposed method, and the accuracy of the welding depth extraction improves by up to 32% compared with the percentile filter.

In this paper, an OCT welding depth extraction method based on LOF and maximum filter is proposed. This is done by first detecting the noise points of the raw OCT data using LOF and then applying the maximum filter to the cleaned OCT data to extract the welding depth. By comparing the welding depth extracted from OCT data with the actual welding depth extracted from the longitudinal cross section, the average error of the proposed method is less than 4%, and the accuracy of the welding depth extraction improves by up to 32% compared with the percentile filter. The proposed method can well improve the accuracy of the welding depth extraction and is more applicable. The laser welding depth monitoring method based on OCT can achieve high measurement accuracy in continuous monitoring and provide quality assurance in industrial production. Further, it will be developed to realize the welding depth control for laser welding.