This paper presents the evolution path of intelligent technologies, from automated monitoring at early stage to modern AI-based decision support, and focuses on analysis of research results and application cases of key technologies, such as neural networks, the Internet of Things (IoT), digital twins and big data. It also presents discussion of the latest progress in application of a new generation of intelligent technologies represented by machine learning in safety management of non-coal mines. Specifically, technologies including deep neural networks, support vector machines, decision trees and random forests have shown advantages in aspects such as personnel safety management and positioning, monitoring and early warning of mine disasters, safe operation of mine equipment, and safety in mine blasting operation. Finally, the development trends of non-coal mine safety management are proposed, in terms of in-depth application and integration of intelligent technologies, compatibility of old equipment with modern intelligent technologies, and integrated management.

To meet the requirements during a 2-year transition period from open-pit to underground mining in Miaochong Iron Mine, its productivity continuity and mining stability were analyzed. Based on ore reserve estimation, Panel ① and Panel ② at the southern end of the roadway at midsection (370 m) of ore body are decided to be an initial mining plan for the transition period. Then, a model including ore bodies, rock formation, open-pit mine and underground roadways is built with FLAC3D software, which is then adopted to analyze the displacement, stress and plastic zone distribution characteristics of the open-pit and roadways at the midsection (370 m) during the transition period. The results show that the ore quantities in Panel ① and Panel ② at the  midsection (370 m)  during the transition period can ensure production capacity requirement, but also is sufficient for subsequent mining. It is found that the overall displacement of the open-pit and roadways at the midsection (370 m) is less than 5 mm, and compressive stress becomes dominant in the stress distribution, which is far lower than the stress resistance of the rock mass. There is no large-scale continuous plastic zone, indicating good stability and small impact by mining disturbance. In practical production, it is recommended that a segmented millisecond blasting should be adopted to achieve a staggered blasting operation  between open-pit and underground mine. Besides, efforts should be intensified in monitoring open-pit slope and underground roadways to ensure safe production during the transition from open-pit to underground mining.

To address high cost of smooth blasting using detonating cords for mine’s roadways, explosive train in blast hole with only one detonator was realized by taking advantage of sympathetic detonation (SD) of explosive. After performing experiments on SD with emulsion explosive in different charge diameters, with different charge weight and other constraint conditions, the SD distances under different conditions were obtained. The results show that SD distances of emulsion explosive with a charge diameter of 27 mm in a PVC tube and blast holes are 10 cm and 90 cm respectively, and constraint conditions have a significant impact on the SD distance of explosives. With the charge weight increasing from 150 g to 300 g, the SD distance is at least 5 cm longer, indicating the SD distance increases with the increase of charge weight. On the other hand, the charge diameter of emulsion explosive also exerts some impact on the SD distance. The SD distance of emulsion with a charge diameter of 35 mm in blast holes is 5 cm longer than that of the emulsion with charge diameter of 27 mm. Industrial experiments on blasting show that the roadway constructed by adopting sympathetic detonation can be well formed, which can not only meet requirements of the subsequent tunnelling in mine but also cutting blasting cost.

To explore the influence of land use change on landslide susceptibility, Liuyang City of Hunan Province was taken for research. An improved infinite slope model was established based on land use data from 2000 to 2020, by using an inversion model of root cohesion and dynamic hydrological correction, and the influence mechanism of land use change on landslide susceptibility was quantified. The model accuracy was verified by comparing six types of parameter combinations. It is found that the model considering spatial heterogeneity of root cohesion and saturated hydraulic conductivity corrected based on land use has extremely high prediction accuracy (AUC=0.8150), presenting a significant improvement compared with the traditional model. Besides, forest land conversion to land for construction significantly increases landslide susceptibility (with a frequency ratio of 1.4269), and decline of vegetation cover leads to reduction in root reinforcement.

In order to select an appropriate deep-sea polymetallic nodule collection method, a technical evaluation system was established based on five criteria, including green, economy, reliability, intelligence and safety (GERIS). Then, as for three kinds of collection methods, including collection with a rotary raking system, or a hydraulic collection system, and a robotic arm collection system, the weights of 22 engineering indicators were determined by adopting the Analytic Hierarchy Process (AHP), and also a quantitative comparison was made among those three typical methods. It is found that collection by a rotary raking system presents better performance in terms of collection efficiency, operation stability and safety, with a comprehensive score of 80.1, while the hydraulic collection system is characterized by simple structure and high efficiency, but is not environment-friendly and intelligent, with a score of 76.3. The robotic arm collection system shows some sort of potential in intelligent development, but is limited by its low efficiency and high mechanical complexity, with the lowest score of 70.2. Overall, the collection by a rotary raking system is more feasible in current engineering applications, and the hydraulic collection system may be banned to use due to strict environmental protection regulations, while the robotic arm collection system still has developmental potential for precise collection in high-value areas in the future.

The typical region of the Yuanma Basin (covering Mayang, Luxi and Yuanling counties) was taken in the research of landslide disaster. Based on the database of detailed survey of geological disasters from 2013 to 2022, landslide events with complete rainfall records were selected for research. Then, significant rainfall factors for landslide, including rainfall on the day of landslide occurrence and during those four days prior to it, and maximum rainfall rate on the occurrence day were selected out by adopting logistic regression analysis, and a logistic regression prediction model for rainfall-induced landslides in the Yuanma Basin was developed and validated by  studying representative landslide cases. The research results indicate that the rainfall-induced landslides in the Yuanma Basin is significantly correlated with rainfall on the occurrence day and during those 4 days prior to the occurrence. This confirms that the period including the occurrence day and those 4 days prior to the occurrence is critical for early warning of landslide in this region. A combined rainfall factors, consisting of R₀  (daily rainfall on the occurrence day), R₁ (rainfall of 1 day before occurrence), R₂  (rainfall of 2 days before occurrence) and R_h  (maximum rainfall rate on the occurrence day), present optimal significance in statistical analysis. The practical cases have validated that this logistic regression model constructed based on this factor combination has a prediction accuracy of 83%. It is concluded that this model can give a prediction with high precision and is of practical value.