To ensure the forming quality and stability of laser powder bed fusion (LPBF), an online monitoring technology must be developed. Using diodes to obtain the radiation intensity of molten pools can accurately reflect their temperature and other important information in real time. It is an economical and efficient method for monitoring the forming process and features low cost, high sensitivity, rapid response, and a small sample size. However, the associated analysis is limited and challenging; therefore, the mapping relationship between the radiation monitoring data of molten pools and the forming process must be analyzed. Process parameters significantly affect the performance of metal additive-manufactured components. Most current production process parameters are not necessarily optimum, which implies that they may cause manufacturing defects. Determining the optimal forming process parameters for each metal via numerous forming experiments is time consuming and labor intensive. Therefore, by analyzing the mapping relationship between the process parameters and radiation intensity of a molten pool, quality monitoring and process adjustments can be realized through online monitoring. This can prevent defects or even failures caused by process parameters and significantly reduce the time, manpower, and financial resources required for forming experiments. Although the radiation intensity of molten pools and the related process parameters have been investigated extensively, a method for adjusting the process parameters directly using the radiation intensity of the molten pool has not been devised. Therefore, the effects of different process parameters on the radiation intensity of molten pools must be investigated to identify and adjust the process parameters.

To determine the mapping relationship between the process parameters and radiation intensity of molten pools, three methods for analyzing the radiation intensity of molten pools were proposed: the deviation integral curve, the probability density function curve, and relative heat-map analysis methods. Three groups of powder bed forming experiments with different process parameters were designed, and the radiation-intensity data of each sample were obtained. After performing data preprocessing via data segmentation, noise reduction, and feature extraction, single-layer and multilayer radiation-intensity data-analysis methods were used for comparative analysis.

As laser energy increases, the temperature of the molten pool increases and the liquid-phase duration is prolonged. The powder around the droplet is easily inhaled, thus resulting in more powder at this position and less powder at the next position compared with the normal condition; therefore, the temperature of the molten pool fluctuates significantly, thus causing the signal obtained by the diode to fluctuate considerably. When the scanning interval is reduced to 0.06 mm, or the scanning speed is reduced to 1000 mm/s, or the laser power is increased to 350 W, i.e., the corresponding laser energy is increased from 93.75 to 281.25 J/mm³, the molten-pool stability is reduced and the fluctuation range of the diode signal increases. Therefore, the corresponding single-layer deviation integral curve fluctuates more significantly , the dispersion degree of the multilayer deviation integral curve increases, the shape of the single-layer probability density function curve is short and round, and the average distribution of the multilayer probability density function curve becomes increasingly wider. The relative heat map shows a high portion of non-green areas.

A molten-pool radiation intensity analysis method was proposed to analyze the radiation intensity of a molten pool under different process parameters from different perspectives, and a unified influence law was obtained. As the scanning distance and scanning speed decrease and the laser power increases to a certan range, that is, the laser energy density increases to a certain extent, the radiation intensity of the molten pool generally increases, the fluctuation degree of single-layer data increases, and the distribution range of multilayer data increases. Our study has realized the recognition of abnormal layer, but the influence of abnormal layer and the measures to avoid it have not been studied, so they should be considered to be explored in the future. Additionally, the effects of metal vapor and splash on the online monitoring of molten-pool radiation intensity shall be considered.