Hybrid Model-Based Soft Sensing Method for Temperature in Laser Soldering Processes

With the miniaturization of electronic products, solder joints on consumer circuit boards have become denser, with sizes reaching the millimeter scale. Accurate temperature measurement is therefore crucial for the quality control of laser soldering processes. However, infrared thermometers measure the surface temperature rather than the internal temperature of the solder paste; in addition, the measured temperature is susceptible to influences of volatilized gases and environmental disturbances. This paper presents a hybrid model-based temperature soft sensing (HMTSS) method for laser soldering processes. This method combines a thermal conduction model (TCM) with a data-driven model. The TCM, which describes the relationship between the internal and surface temperatures of solder paste, is derived from the principle of energy conservation. The data-driven model, which is built from extensive experimental data, corrects the prediction errors of the TCM and thus improves the accuracy. The data-driven model follows a staged modeling approach in which a support vector regression (SVR) model serves as the base model to address the errors in the heat transfer model and a random forest (RF) model corrects the output of the base model. The data-driven model leverages the strong generalization capabilities of the SVR and the robustness of the RF, achieving highly accurate predictions. The effectiveness of the HMTSS method was validated experimentally. The results demonstrated that the proposed method predicts the internal temperature of the solder paste with errors within ±10°C, indicating relatively high accuracy.