Abstract: The phosphorus and chlorine elements in boiler water may cause pitting corrosion, which can form corrosion pits in local areas of the weld metal, gradually eroding and thinning the wall thickness of the weld metal. Once the weld metal is penetrated, it will cause medium leakage and lead to the failure of the boiler system. Therefore, a method for detecting welding failure of boiler short tubes under low load conditions is proposed. Using laser-induced breakdown spectroscopy (LIBS) technology to perform high-quality spectral measurements on the welding joints of boiler short tubes, optimizing the measurement architecture to avoid the influence of oxide and deposition layers on the detection effect. In response to the spectral characteristic fluctuations caused by the non-uniformity of the metallographic structure inside the boiler short tube, spectral correction techniques including Savitzky Golay convolution differentiation and least squares fitting are adopted to improve the accuracy of spectral data. And use the K-SVM-RFE algorithm to select spectral data features, combined with the decision function of linear kernel SVM to achieve welding failure detection of boiler short tubes under low load conditions. The experimental results show that under low load conditions, the failure detection of the design method only results in 3 detection errors, and the sensitivity is above 0.95, which can achieve the failure detection of boiler short tube welding under low load conditions.