The structural stability of water diversion tunnels is paramount for ensuring the safe operation of hydropower stations, with regular inspections being a critical measure to maintain engineering safety. However, existing monitoring methods are often plagued by inefficiency, incomplete defect identification, and low accuracy in quantitative characterization, primarily due to challenges such as poor lighting, enclosed spaces, and inaccessible inclined shafts. By leveraging technologies and methods such as sensing, robotics, deep learning, and multivariate statistics, an integrated technology for tunnel safety, namely "in-situ monitoring-mobile inspection-defect quantification-risk identification", has been established and applied in the water diversion tunnel technology of the Houziyan Hydropower Station on the Dadu River. The results show that from 2013 to 2021, under in-situ monitoring, the displacement of tunnels 1^# and 3^#, the stress of anchor rods, the opening degree of joints, and the osmotic water pressure first presents a sharp change, then gradually slows down, and finally tends to stabilize. The defect prediction model based on the least squares fitting method has a better fitting effect (R² ranges from 0.60 to 0.91), and can predict the changing trend of each defect over time. The inspection results of the robot show that the leakage, cracks and other defects of Tunnel 3^# are mainly distributed on the left and right waisides and the top of the tunnel. Among them, the leakage through cracks is the most prominent. There are a total of 107 cracks and leakage points with obvious leakage phenomena. Multivariate constrained ranking analysis indicated that the defects in Tunnels 1^# and 3^# of the Houziyan Hydropower Station remain within the "relatively stable" range, with seepage, seepage pressure, and cracks identified as key risk factors affecting tunnel safety. Robotic inspection technology proved to be both safe and efficient, unaffected by environmental conditions or subjective biases, and capable of capturing data that in-situ sensing methods could not. The study achieves a transformative shift in tunnel safety management from "passive emergency response" to "active prevention and control" through its integrated "Monitoring-Inspection-Identification-Evaluation" technological chain. The research outcomes provide effective technical solutions and scientific theoretical references for safety inspection and risk assessment of tunnels both domestically and internationally, demonstrating significant academic research and practical engineering value.