Intelligent Recognition Of Rock And Coal Failure Precursors Using Infrared Radiation

The damage and fracture of rock are the fundamental causes of coal mine dynamic disasters such as coal and gas outbursts,rock bursts,and mine water inrush.These are the basic and common scientific problems in rock engineering.Accurate and effective rock damage and fracture monitoring can provide reliable precursor information for coal mine dynamic disasters.It is an important basis for monitoring and early warning of surrounding rock fracture and water seepage（water inrush）in the mining face.The rock failure precursors information can be captured effectively using infrared radiation（IR）.Furthermore,the changes in IR characteristics can assess the rock damage and fracture under loading in a better way.However,most current studies focus on the IR variation characteristics of the loaded rock surface but do not establish a quantitative relationship between the IR information of the rock surface,its internal damage,and fracture characteristics.This flaw restricts the application of IR in effectively monitoring mine disasters and establishing early warning.Therefore,it is essential to determine early warning signals for better monitoring of mine disasters for the safe and efficient execution of mining activities.In this thesis,uniaxial and shear experiments of sandstone under different loading rates are carried out using the self-designed multi-parameter monitoring system of loaded rock.The characteristics of acoustic emission（AE）and surface IR of loaded sandstone are explained,and the internal relationship between IR,internal damage,and rock fracture is established.A predicted model for the different stages of rock under loading using IR and Rock Convolutional Neural Network is proposed.The findings of the research are presented as follows:（1）AE and average infrared radiation temperature（AIRT）characteristics were studied under different shear angles.The AIRT behavior was studied in shear and tensile cracks.The findings revealed that shear stress and strain decrease with the increase of water content while shear stress is inversely proportional to the preset shear angle.The AE energy is relatively low in water saturation compared to the dry sample.However,a large proportion of the accumulated AE energy was released near peak stress in dry samples.The AIRT in dry rock almost linearly decreases with loading time and abruptly increase at the failure point.Whereas with an increase in water contents,a linear trend is moving to a nonlinear trend,at crack growth or generation,the AIRT increased or decreased;The ratio of AF（counts divided by duration,k Hz）and the RA（rise time divided by peak amplitude,ms/V）is proposed which indicates that shear crack is governing in the failure process under shear loading.Its intensity increases with water content and ratio,while tensile cracks decrease with the increase in water content.Furthermore,AI models such as random forest（RF）,decision tree（DT）,and support vector machine（SVM）are used for shear and tensile crack based on AIRT.The model’s efficiency evaluation revealed that all models show higher accuracy（R²>0.90）,but the RF show remarkable accuracy.The total shear and tensile crack are 70468 and11112,respectively.（2）The AE characteristics,AIRT,cumulative AIRT,fractal characteristics of AIRT,and infrared radiation thermograph（IRT）were used to predict the different stages of stress-time curve for sandstone.Using IR characteristics for the prediction of shear and tensile cracks more conveniently.Results show that stress-strain curve is divided into four stages based on the AE characteristic:crack closure,elastic deformation,stable crack propagation,and unstable crack propagation.In crack closure,the AIRT drastically decreases,and cumulative AIRT increases;In elastic deformation,AIRT slightly decreases,and cumulative AIRT increases and reaches the highest value.During stable crack propagation,the AIRT remains almost constant,and cumulative AIRT steeply decreases,while in unstable crack propagation,the AIRT slightly decreases,while cumulative AIRT gently decreases.The fractal characteristic in the first two stages increases,and in the stage,the rate of increase is less than in the previous two stages,while in the last stage,it decreases drastically.Furthermore,stress-strain curve stages were also predicted using AIRT and critical slowdown theory（CSDT）.The stress level at each stage was strongly correlated（average）with the actual stage’s（defined by AE）stress level.It is worth mentioning that the prediction of stages through a non-destructive test was the first time predicted.The convolution neural network of rock models was proposed to predict different stages of stress curves（under different water content）using IR temperature images.The performance shows that the proposed model gives high accuracy overall（98%）.（3）The effects of water contents on sandstone dilatancy point under loading in the presence of IR were investigated.Furthermore,the dilatancy（crack damage）point index was predicted from IR data using three computing techniques;artificial neural network（ANN）,RF,and k-nearest neighbor（KNN）.The performance of all techniques was evaluated using performance coefficient（R²）and root means square error（RMSE）.Results revealed that water content affects the different stages of the stress-strain curve,resulting in a reduction of elastic deformation and stable crack propagation stages.The rate of IR variance at crack initiation and dilatancy point increases,and the corresponding stresses decrease with water content.The absolute dissipation strain energy rate suddenly changed at the dilatancy point.Therefore,this can also be used as a precursor index for rock failure.The dilatancy point index was predicted using different models,which shows that KNN model predicts its value effectively compared to ANN and random forest regression.（4）The ratios of elastic to dissipation energy(K_ED)and elastic to the total energy(K_ET)were proposed to predict the early failure point（EFP）of sandstone under different water contents.Results revealed that K_ED and K_ET give EFP at the same time for sandstone with water contents;0%,0.991%,2.136%,and 3.109%,and the average time of EFP ahead 208.0s,250.0s,265.8s,and276.9s than rock failure,respectively.Furthermore,the proposed K_EDand K_ETwere predicted using ANN.The ANN models’efficacy was evaluated using the R²and RMSE.The findings revealed high R² and low RMSE for K_ED and K_ET sandstone with different water contents.（5）The rock failure precursor information was improved using CSDT based on variance and autocorrelation for IR indexes,i.e.,variance of infrared radiation temperature（VIRT）and variance of differential infrared image temperature（VDIIT）,under loading on coal samples at different loading rates.Results revealed that the sudden and significant critical transition（increase）in IR indexes variance and autocorrelation coefficient could be used as an early warning sign of coal.The ratio of early warning point（inflection point）to peak stress and time for both IR indexes is approximately 91%and 80%,respectively,for 0.1 mm/min.This ratio decreases with an increase in loading rates.The effect of window length on variance and autocorrelation of both indexes has less or no influence on the inflection point range but affects the fluctuation amplitude.The variance amplitude is inversely proportional to window length,while autocorrelation is directly proportional;The precursor information based on autocorrelation stress level decreases（about 3.4%for VIRT and 4.2%for VDIIT）with an increase in loading rate and permits precursor identification in a detailed and effective manner,which gives a significant guideline for coal pillar instability prediction.There are 83 figures,23 tables,and 218 references in this dissertation.