Mine fire is easy to lead to heavy casualties due to strong sudden occurrence and difficulty in disaster relief and fire fighting.Early and rapid fire identification and emergency control is the key to prevent serious accidents.At present,the underground fire detection method is to monitor the early fire characteristics by placing fire characteristic gas sensor at a fixed location,which is relatively simple and has a large lag.Once a fire occurs in the mine,the intensity and scope of combustion will be rapidly expanded under the strong interaction between ventilation and thermal power,which greatly threatens the safety of personnel.In addition,most mines have not realized the intelligent control of ventilation facilities,unable to timely control the spread of disasters leading to heavy casualties.Based on this,this thesis combines the deep learning-based visual detection technology with multi-sensor fire monitoring technology to detect mine fires,and uses PLC control technology to build an automatic fire emergency control system to realize remote emergency linkage control during disasters.The main work of this thesis is as follows:Based on the one-stage object detection algorithm YOLOV5,the model training process was optimized by using the label smoothing strategy,and the data enhancement method was used to further enrich the data set.By comparing with the original model,it was found that the adopted strategy had an optimization effect on the training accuracy.Four attention mechanisms,SE,CBAM,ECA and CA,were added to different locations of the backbone network and network neck.It is found that the training effect of adding attention mechanism to the backbone network is better,and the ECA attention mechanism has the best effect compared with other attention mechanisms.In order to meet the detection requirements of small targets in the YOLOV5 detection network,the detection feature map of very small targets is added to the original network,and the final output feature map size is 120×120.It is proved that this method can detect more small targets under the same conditions.Finally,the binocular camera was combined with YOLOV5 to realize the accurate location of fire source point and complete the visual detection of fire.The numerical simulation method is used to establish the fire evolution model of belt roadway in Zhuanlong Wan.Through the simulation of 6 different wind speed conditions,it is found that with the increase of wind speed,the spread distance of the belt also increases,and the heat release rate of the roadway fire shows an upward trend.By analyzing the migration law of smoke,CO and temperature,the best spatial locations for detecting the early characteristics of fire under different wind speed conditions are obtained.The influence of visibility and CO on personnel escape in roadway with different wind speed was analyzed.The full-size modeling of belt roadway and western mining area at the back of Zhuanlong Wan Ⅱ-3 coal mine was carried out to study the law of smoke flow migration after adopting the fire emergency linkage control method during the disaster.The fire protection zoning of Zhuanlong Wan Coal Mine was divided,disaster simulation and network calculation were carried out in different regions,and the influence of disaster on the whole ventilation network was analyzed.It was found that the air volume demand of each site could still be satisfied after the disaster.In order to ensure the timeliness of disaster control,the visual detection technology and sensor monitoring technology are used to detect the fire characteristics comprehensively,and the remote emergency linkage control method of fire is proposed,and the automatic control system of ventilation facilities is constructed.Adopt PLC as the core of the main controller,and add local remote control and touch screen and other trigger ways.The basic operation requirements of ventilation facilities and disaster control requirements are realized based on a variety of logic controls.The controller adopts spare batteries for redundancy design to ensure that the controller still has the corresponding control function after the disaster power failure. |