As one of the important contents of bridge health detection， crack detection reflects the stress and damage state of bridge structure. The traditional bridge crack detection is mainly based on human eye recognition， of which efficiency and accuracy are both low. Moreover， the human eye recognition has the following problems such as effected greatly by illumination， incapability to detect in some high-altitude positions like bridge towers and high piers and strong subjectivity. In recent years， many scholars at home and abroad have developed many bridges crack detection equipment based on digital image technology to solve the above problems， such as bridge detection vehicles equipped with high-definition cameras， drones， and climbing robots. Meanwhile， the efficient and high-precision crack detection algorithm is the basis of crack detection. How to balance the detection speed and accuracy has always been one of the hot issues studied by many scholars. In this paper， the bridge crack detection equipment based on digital image technology， the platform and calibration method of camera， preprocessing algorithm， traditional detection algorithm， deep learning algorithm， crack feature calculation， image stitching algorithm and three-dimensional output and monitoring of cracks are reviewed. In addition， summaries to deficiencies in the study and prospects the bridge crack detection method， crack three-dimensional expression， crack monitoring and management， bridge stiffness loss evaluation and early warning for the future development trend.

A novel bamboo-like hexagonal gradient hierarchical multicellular tube （BHGHMT） was designed. A theoretical solution of the mean crashing force for hexagonal single-gradient hierarchical multicellular tube （HSGHMT） is derived by using the super-folded element method， and the experimental study of the square tube is carried out. The accuracy of the finite element numerical model is verified by using the theoretical solution and the experimental results. The crashworthiness performance of BHGHMTs with different wall thicknesses was analyzed using the validated numerical model. Finally， the crashworthiness performance of the BHGHMT and the conventional square tube （ST）， hexagonal tube （HT）， and HSGHMT under the same mass， the same initial peak force and the same energy absorption were compared. The research results show that the proposed BHGHMT can effectively reduce the initial peak force and improve the crushing force efficiency. The crashworthiness performance of the BHGHMT is better than that of the conventional ST， HT and HSGHMT under the same mass， the same initial peak force and the same energy absorption. The research results can provide a reference for the innovative design of gradient hierarchical multicellular tubes.

The bogie area of high-speed trains is one of main sources for aerodynamic noise. So far， it is difficult to describe the characteristics of aerodynamic noise sources of high-speed trains， there are few effective sound source identification methods. Using the characteristics of the sound source dominated by dipole sound sources in the high-speed train bogie area， the aerodynamic sound source is equivalent to a collection of countless spherical sound sources. Based on the relationship between sound radiation and sound source， sound source and flow field physical quantities according to the relationship， combined with the fluid numerical simulation， a method for identifying the sound source of the high-speed train dipole was established， and the sound source was identified by focusing on the bogie area of the lead car. At the same time， based on the vortex sound theory， the relationship between the intensity of the dipole sound source and the multi-physics of the flow field was established， and the essence of the sound source generated by the flow field was analyzed. The research shows that the concentrated position of the dipole sound source is mostly the position where the windward side airflow and the wall surface strongly interact， and the collision and separation of the airflow and the wall surface are the main reasons for the generation of the dipole sound source， and the change of vorticity in this area is dual. The intensity of the polar sound source has the greatest influence， and in different regions， the vorticity components in different directions play a leading role.

A new type of mobile chassis with H-type balancing rocker mechanism for road adaptive inspection robot is designed for outdoor inspection in the petrochemical industry， which adopts a new “4h” symmetrical structure to realize four-wheel drive and steering operation of the chassis. Based on the introduction of the overall structure of the chassis， the design principles of the chassis drive mechanism and steering mechanism are analyzed， and the theoretical models of the chassis Ackermann 4WS steering and in-situ steering are established to provide a theoretical basis for the motion control of the robot chassis. The theoretical analysis and simulation of the unilateral obstacle-surmounting process of the robot chassis were carried out， and the simulation comparison with the obstacle-surmounting process of the common integrated chassis was conducted to study the driving stability of the chassis during the obstacle-surmounting process. Finally， the steering mode experiment and unilateral obstacle surmounting experiment of the chassis prototype were conducted. The results show that the mobile chassis can realize the switching of drive and steering modes， and the chassis has a smooth motion attitude when conducting unilateral obstacle surmounting， with good road self-adaptive ability and obstacle surmounting performance.

Considering that the cylindrical shell structure is prone to produce the petal-shaped mode shape， a method of vibration suppression is proposed to attach multiple piezoelectric shunt patches with equal-angle. Based on the ANSYS platform， the coupling between the cylindrical shell and the piezoelectric patches structure is realized， and the parameter input method of the electronic components in the shunt circuit and the coupling process between the piezoelectric patches are given. A case study is carried out on a cylindrical shell with four piezoelectric shunt patches attached with equal-angle as the object， and the rationality of the proposed finite element modeling method and the vibration reduction effect of the piezoelectric shunt damping patches on the cylindrical shell are proved by experiments. Based on the analysis of the influence of parameters， it can be found that there is an optimal resistance value for the resistance shunt circuit to achieve the optimal damping effect and it is beneficial to reduce vibration when piezoelectric patches are attached to the region with relatively large deformation and increasing the size of piezoelectric patches.

In order to solve the problem that mechanical weeding between rice plants depends on the alignment of the machine and the seedling belt， an inter-plant weeding machine without avoiding seedlings was designed. The rear weeding component adopts the rake method to remove weeds. It controls the start and stop of the vibration operation in real time according to the front row recognition result. A tactile perception method was proposed to establish a rice plant posture recognition system. Firstly， the data of the roots of the rice plants on the front row of flexible brushes was obtained through the changes in the voltage data caused by the changes in the air pressure by the air pressure sensor. Then， the kurtosis feature and the mean feature in the data were extracted. Finally， the feature vector was substituted into the support vector machine to construct a rice plant posture recognition model. In this paper， a hydraulic motor and rear weeding components are used to form a vibration system， ensuring weeding components can move laterally. When the sensing system detects that the rice plant is fell down， the vibration system will stop working. The field test shows that the recognition accuracy of rice plant lodging is 83.11%. When the sensing system is not working， the weeding rate of the vibration system is 58.11%， and the seedling damage rate is 21.39% .The weeding rate with neither the sensing system nor the vibration system working is 27.89%， and the seedling damage rate is 4%. The weeding rate of the operation with sensing system and the vibration system at the same time is 56.17%， and the seedling damage rate is 5.83%. Based on the test results， the simultaneous working mode of the sensing system and the vibration system is more suitable for the removal of inter-plant weeds in the paddy field.

The precision of traditional precision positioning stages are limited due to mechanical friction， so it is difficult to meet the motion requirements of large stroke and high precision. In this paper， a single drive rigid-flexible coupling positioning stage combined with macro- micro composite structure was proposed. Compared with the other macro-micro composite stage， it greatly simplified the design of mechanical structure and motion control system. Based on the floating coordinate method and finite element method， the static and dynamic model of the single drive rigid-flexible coupling positioning stage was established. The analysis results of these models were compared with the static analysis results of ABAQUS finite element software and the analytical solution of the simplified dynamic model， the maximum relative errors were 1.6% and 3.72% respectively. It is proved that this theoretical model has high prediction accuracy， which can provide parameters for the optimal design and precision motion control of the single drive rigid-flexible coupling stage.

In the present study， 3 mol% Yttria-Stabilized Zirconia （3YSZ） powder was used as the base material doping with conductive phase-carbon nanotubes （CNT） to conduct low-temperature sintering at 200 ℃， and simultaneously assisting AC electric field and pressure. The effects of carbon nanotube concentration on the phase composition， microscopic morphology and ionic conductivity of sintered materials were mainly discussed. The results showed that the composites were sintered uniformly and densely at this temperature， and the ionic conductivity was 0.20 μs/cm when the concentration of carbon nanotubes was 2%；When the concentration of carbon nanotubes was 3%， the ionic conductivity 0.83 μs/cm could be reached；When the concentration of carbon nanotubes was increased to 4%， the ability of the material to withstand electric field strength and current density was further improved， and the initial voltage of flash sintering was reduced， but the conductivity was 0.6 μs/cm at this time， meantime， the material could be sintered at room temperature.

In this paper， three-point bending tests of T700 carbon fiber composite multi-directional laminates with different thicknesses are carried out under two stress ratios and three stress levels， and the static and fatigue properties of three-point bending are analyzed. The morphology after damage was observed in different ways. The results show that the increase of material thickness increases the flexural strength and modulus of the specimen， and the three-point bending fatigue life of composite multi-directional laminates is proportional to the material thickness， and inversely proportional to the test stress ratio and stress level. The final fatigue failure degree of the sample is related to the fatigue life， the higher the number of fatigue cycles， the more severe the damage in the transverse and longitudinal directions of the plywood. At the same time， the greater the thickness of the sample， the more obvious the delamination phenomenon， and the phenomenon of fiber pull out.

In this paper， situational awareness of drivers engaged in different types of non-driving-related tasks （NDRTs） is conducted for takeover readiness of L3 automated vehicles. Simulation testing scenarios are built based on driving simulator platform. Then， the situational awareness readiness was discussed and analyzed through the situational awareness questionnaire and the driver's eye movement behavior measurement. Experimental results show that as NDRTs occupy more sensory modalities， the driver's attention will be less focused on the driving task， and the attention to the elements of the driving environment will be correspondingly reduced， which has a negative impact on the recovery of situational awareness.

In order to describe the car following stability of the mixed traffic flow under the condition of Connected Automatic Vehicle mixing， reveal the changing process of mixed traffic flow stability， analyzed the stability of HDV and CACC traffic flow， introduced the mixing rate of CACC， proposed the following model of mixed traffic flow， and derived the stability criteria of mixed traffic flow stability. Analyzed the effects of speed and mixing rate on the stability evolution of mixed traffic flow under single and combination conditions respectively. The results show that： The stability of mixed traffic flow depends on the stability of a single traffic flow and the mixing rate. The critical speed values of the following stability of HDV and CACC single traffic flow are 19.65 km/h and 18.01 km/h， respectively， the speed is positively correlated with the stability of mixed traffic flow. The effect of mixing rate on the stability of mixed traffic flow is related to the speed condition. Under the combined conditions of the two factors， the stability changing trend of mixed traffic flow is diversified.

In order to promote the accuracy of risk level identification， to solve the real-time risk control and emergency treatment problems of urban rail transit system， an improved feature selection algorithm （Im-F-score+XGB） on filtering features of the risk factors for emergencies was proposed. Through the analysis of the basic data of urban rail transit emergency， the feature importance degree of each risk feature was calculated， the influence degree of different features on the risk of emergencies was excavated， and the important features for determining the risk level of emergencies was obtained. Besides， the cyclic multi-time window scanning method and the weighted cascade residual forest model were used to obtain the mapping relationship between the emergency risk grade and the feature of risk features， and an improved emergency risk level identification model （Im-F-GCF） was established. Compared with RF， HGBDT， GCF and LightGBM models， the validity of the proposed model is verified.

To obtain the dynamic response characteristics of long-span continuous beam bridges under coherent excitation of the bridge deck， a comprehensive analysis was conducted based on the principle of virtual work. Firstly， a dynamic model of a five-axle heavy-load vehicle was established. Then， the bridge's dynamic equation was derived using the modal superposition method， a wheel-deck coherent excitation model was established by considering the geometric and mechanical coupling between the wheels and the bridge. The coupled vibration analysis system of the long-span continuous beam bridge was implemented using Matlab software. Subsequently， the impact of different levels of coherent deck excitation models on the dynamic response of the bridge was investigated. The findings revealed that when the influence of bridge deck irregularity was ignored， the impact coefficients of the bridges were smaller than the standard values. However， as the condition of the bridge deck deteriorated， the dynamic response of the representative cross-section of the structure increased significantly. Moreover， it was observed that the completely coherent excitation model had a greater influence on the dynamic response of the bridges compared to the partially coherent excitation model， exceeding the standard values. The maximum deviation of the deflection amplitude was approximately 7.7%， with a maximum deviation of bending moment amplitude of around 20%. The relative errors in deflection impact coefficient ranged from 4% to 34%， while the relative errors in bending moment impact coefficient ranged from 2% to 43%. Additionally， it was noted that the impact coefficient of the bridge initially increased and then decreased with an increase in vehicle speed. Finally， an envelope diagram illustrating the deflection and moment impact coefficients of the long-span continuous beam bridge was provided. These research findings offer valuable insights for the analysis and study of the structural dynamics and response of continuous beam bridges.

The carbonization block is a new type of block， which is developed on the basis of the carbonization method of reactive magnesium oxide （MgO） cement， in which the reactive MgO cement is firstly mixed with aggregates and then CO₂ is injected for carbonization to achieve the purpose of improving the strength. The indoor tests of unconfined compressive strength， X-ray diffraction， electron microscope scanning， and thermogravimetric analysis were employed to study the effect of carbonization humidity on the strength and carbonization products. The durability of carbonation block was tested to explore its engineering performance. The results show that the strength of carbonization block is higher than that of ordinary cement block. After 3 days of carbonization curing， the strength of carbonization block can reach 70%~80% of that during 28 days curing， and the strength of block with carbonization humidity of 60% is the best. Carbonation blocks have a strong durability and the durability of carbonated blocks with carbonation humidity of 60% is significantly higher than that of humidity 40% and 80%. The carbonization products of reactive magnesium oxide are magnesite carbonate， brucite and nodular magnesite. The carbonization products in the carbonized block with carbonization humidity 60% are the most dense.

A track irregularity simulation method based on improved triangular series method based on wavelet theory was proposed. Combined with the established 7-rigid-body 42 degrees of freedom vehicle dynamics model and wheel-rail contact model， a dynamic simulation model of variable gauge train was built based on SIMPACK. The suspension parameters and wheel tread were optimized by parameter experiment， and the dynamic performance of the modified gauge bogie was predicted， the critical speed of the vehicle is far higher than the design requirements of 400 km/h， stationarity index reach the standard of secondary vehicle stability level， curve through sexual evaluation indexes is lower than the safety limit， developed by variable gauge bogie train has good dynamic performance.

In order to provide new thoughts and new methods for traffic planning and management of small and medium-sized cities， the travel characteristics of residents in small and medium-sized cities was analyzed. Considering the heterogeneity of residents' travel mode choices， a travel mode choice model of based on random parameters Logit was constructed. Factors of psychological latent variables such as the requirement of comfort were introduced， and the structural equation model was used to calibrate the parameters， which reflects the relationship between latent variables and index variables. The results show that the prediction accuracy of the stochastic parameter logit model calibrated by structural equation is higher than 95%， which indicates that the structural equation model has good parameter calibration effect， and the constructed stochastic parameter logit model has good fitting degree. In the utility function of the model， the coefficient of travel time is a random parameter， which is related to the individual characteristic variables， travel characteristic variables and psychological latent variables of travelers. Among them， the walking time to the bus stop has the greatest impact on the travel utility， and the value of time decreases by 48 yuan for each additional unit. In addition， the marginal utility of different travel modes on driving time is different. The marginal utility of bus turns from increase to decrease after 45 min. Therefore， the random parameters Logit model considering the heterogeneity of individual psychological latent variables can describe the actual travel mode choice behavior of residents in small and medium-sized cities， and has a stronger explanation for the mechanism of travel characteristics.

In a mixed traffic environment composed of connected and autonomous vehicles （CAV） and human-driven vehicles （HV）， a CAV free lane change dynamic control model is proposed to improve the running state of the expressway weaving area. Firstly， by introducing driving style parameters and optimizing HV multi-level lane changing strategy， the HV numerical simulation model of expressway weaving area is constructed， and the constructed model is verified according to different lane changing ratios and lane changing position distribution. Combined with CAV permeability and parameters of lane changing， a CAV dynamic control model of free lane change is proposed. Finally， the influence of CAV dynamic control model on traffic efficiency and safety is analyzed under the condition of CAV permeability random distribution. The results show that the dynamic control model can further improve the traffic efficiency and safety in the weaving area under the condition of low CAV permeability， and the improvement effect of the dynamic control model gradually decreases with the increase of permeability.

In order to diagnose the traffic situation in confluence area from the perspective of traffic infarction， a diagnosis model for traffic situation is proposed with the traffic breakdown probability as the traffic infarction indicator. Based on the proposed model， on the one hand， the satisfaction conditions of model parameters were explored， on the other hand， the traffic situation was investigated with the control variable method and the calibrated model， and the analysis results were discussed and control strategies were proposed. Results show that it is difficult for the traffic flow to leave the confluence area in time when v is small， and P increases when q or r continues to increase. The traffic flow can leave the confluence area in time when v is large， and P is relatively small， but P increases with the continuous increase of q or r. When the traffic flow is known， proper adjustment of v can help reduce P.

Nonlinear dependence of multiple control monitoring point failure exists in bridge member， Considering the influence of this nonlinear dependence on bridge member reliability. In this paper， firstly， Bayesian dynamic model （BDMs） is introduced to build the dynamic prediction model of monitoring data. Further， Gaussian Copula-BDMs is given with the fusion of BDMs and Copula theory， which is used to solve the structural reliability. And the solved reliability is compared with that calculated without considering the dependence among bridge member control monitoring point failure. Finally， an example is given to illustrate the feasibility of the proposed models and approaches. The research results will provide the theoretical foundation for the reliability analysis of structural system.

In view of the problem of serious disease in the western saline soil region， this paper selects a typical environment in the western saline soil region to carry out natural exposure test of reinforced concrete， and simulates the local corrosion environment to design indoor salt spray dry and wet cycle acceleration test. The corrosion current density and concrete damage degree were used to characterize the deterioration laws of reinforcement and protective layer. Based on Weibull distribution function， the deterioration model of reinforced concrete under the two environments were established， and the competitive failure analysis was carried out. Finally， the equivalent relationship between the two environments were established by using the equivalent condition of reliability and the concept of equivalent. The results show that the deterioration law of reinforced concrete under in indoor acceleration is similar to that under natural exposure， and the laws are the result of the interaction of strengthening and deterioration effects. The Weibull distribution function can effectively describe the deterioration process of reinforced concrete under indoor acceleration and natural exposure environment. The time of internal reinforcement deterioration to median life is slightly earlier than that of the concrete protective layer. Indoor salt spray dry and wet cycles accelerated deterioration of reinforced concrete for 1 day can simulate 12.7 days of natural exposure deterioration in saline soil region.

COMSOL Multiphysics finite element software was used to establish a three-dimensional grouting model of high-speed railway ballast-track subgrade. Taking grouting pressure， grouting depth， grouting downward inclination angle and grouting pipe spacing as the main grouting parameters， a numerical simulation study was conducted. The results showed that： The upper surface flux can be used to characterize the amount of grouting. In the process of grouting， the amount of surface mud spillover is significantly positively correlated with the grouting pressure. When the grouting pressure increases by 0.1 MPa， the flux linearly increases by about 0.05 kg/m²·s. The surface flux decreases with the increase of grouting depth， downward inclination angle and grouting pipe spacing. Grouting pressure has a significant effect on the grouting effect. The combined construction method of sleeve valve tube section and upper cover layer is adopted， which can effectively control the mud spillover problem.

The remote sensing image change detection model which is based on multi-scale fusion is proposed to accurately identify the change region of the bi-temporal remote sensing images. First， a multi-scale input pyramid is constructed in the feature extraction stage to receive multi-layer receptive fields and enhance the perception of all information. Then， in order to make a tradeoff between locating the changing area and mining details， the multi-scale calculation is carried out for deep difference features. Finally， the semantic change information can be identified and retained to a great extent by integrating the different feature results of the network. The experimental results show that the proposed model has good performance in both subjective evaluation and objective indexes.

In few-shot image classification tasks， when the model measures similar images of different classes， due to the lack of attention to local important features of the sample and the difficulty in capturing the subtle differences between similar images， the classification boundary between query samples and correct class prototype is fuzzy. Based on this， this paper proposes a Channel Attention Bilinear Metric Network （CABMN）， which firstly increases the attention of the model to the local important region of the image， and then uses the bilinear Hadamard product operation to mine the deep second-order feature information of the important region， so that the model can locate the local key region of the image more accurately. Comparative experimental results show that the proposed CABMN has improved the classification performance on all datasets， especially on the fine-grained datasets CUB-200-2011 and Stanford-Cars， reaching 86.19% and 81.51% classification accuracy.

In order to improve the performance of event detection task， this article redefines this task as a prompt paradigm. This paradigm uses question-answer pairs to transform event detection into machine reading problems. A pre-training model called WLBert-BiGRU is applied to predict the event triggers in QA pairs. The model uses Weight-Layers strategy to enrich the semantic representation ability of Bert model， and uses Bi-GRU to strengthen the prediction ability of the model to the event triggers. The proposed method is evaluated in ACE2005 data set， the results show that the F1 scores in event trigger recognition and classification have reached 78.1% and 75.1% respectively， which is 4.18% and 4.3% higher than the existing work.

In this paper， a control strategy integrating lateral stability and rollover prevention of four-wheel independent drive electric vehicle was proposed in the framework of model predictive control （MPC） to coordinate the lateral， yaw and roll behaviors of the vehicle. Firstly， to ensure the accuracy of the prediction model， the nonlinear tire lateral force model was utilized to predict the future lateral， yaw and roll states of the vehicle. Then， the controller was designed based on the model predictive control method， and multiple control requirements such as improving handling performance， lateral stability， preventing rollover， ride comfort and safety were integrated into the controller. Finally， the simulation results show that the proposed controller can effectively improve the handling stability and lateral stability of the vehicle in high-speed steering， and effectively reduce the roll risk. For the situation of high roll risk caused by the increase of vehicle mass and height of center of mass， even if the controller faces unknown vehicle changes， it can timely and effectively adjust vehicle attitude to prevent vehicle rollover and ensure driving safety.

The SCR system is an important component of the emission aftertreatment systems for diesel engines and is primarily responsible for reducing NO_x emissions. With increasingly stringent emission regulations， the control technology based on precise models for SCR systems is becoming an inevitable choice. During the working process of SCR systems， the model parameters will be shifted due to aging， wear and other effects. In particular， catalyst sulfur poisoning can cause a severe drop in the maximum ammonia adsorption capacity， which can lead to the failure of the SCR system. In this paper， a stepped-up chaos optimization algorithm with single search algorithm （SCOA+SSA） is proposed to solve multi-parameter online identification for SCR systems. This method can simultaneously identify eight key parameters in the chemical reaction kinetics of the SCR system. The experimental results show that this method can identify new system parameters in real time when the maximum adsorption capacity of ammonia decreases in steps， that as compared to the traditional SCOA method， the identification accuracy of the proposed SCOA+SSA method is improved by 5.44% with a computational time penalty of 2.9%.

The inclination sensors based on MEMS accelerometers has been widely used. However， it is well known that a large temperature drift exists in MEMS accelerometers. In this paper， the temperature compensation method based on the normalization of diagonal elements of the transformation matrix was proposed. Both the transformation matrix accounting for the sensitivity， non-orthogonality and misalignment and the other matrix accounting for the bias were obtained with the conventional six-position method. The problem that the matrix related to the sensitivity inconsistence can't be separated from the matrix related to the non-orthogonality was overcome through the normalization of the diagonal elements of transformation matrices obtained under the different temperatures relative to the corresponding elements in the transformation matrices obtained under the room temperature. As a result， the temperature drift of the sensitivity and bias of MEMS accelerometers was compensated. Under the testing temperature of -20 ℃~50 ℃， the precision of the inclination sensor was obviously improved after compensation.

Aiming at the problems of silage baling and film wrapping link separation， large supporting power， an silage baling and wrapping film machine with electric drive was designed. By analyzing the working principle of the machine and the operation process of the core components， the main parameters such as the power of feeding and baling device， the linear speed of conveyor belt， the opening angle of the baling chamber， and the time required for 2~3 layers of film wrapping were determined， the reliability simulation of the frame at full load and the performance test of the whole machine were carried out. The results show that the frame maximum stress was 116.4 MPa and the maximum deformation was 0.71 mm under full load condition， which meet the mechanical properties of the material. The bundling rate was 97.8%， the regular bundling rate was up to 100%， the bale density was 148.2~162.5 kg/m³， and the straw bale resistance rate was 84.0%. The whole machine runs smoothly and reliably with small vibration， the low straw bale resistance rate may be caused by the straw being too uniform and short， and the twine was not complete， which can be optimized in the later crushing treatment and twine mechanism.

In order to solve the problem of arch lifting and leakage of hole-type seeder in high-speed operation （more than 4 km/h）， a new method of high-speed seed filling with arrangement and state-of-motion adjustment guided by limited gear-shaped side space （seed holding space） based on stress of the seed group was proposed， and the corresponding seed tray was designed. By establishing the kinematic model of the filling process of cotton seeds in the seed holding space， the mechanical analysis of the cotton seeds with arrangement and state-of-motion adjustment is carried out， and how to quantitatively separate the seeds step by step through the three stages of semi precision filling， self-adaptive state-of-motion adjustment and migration， and rotating quantitative filling is described， so as to avoid the arch lifting and leakage in the filling process. The box Behnken central combination method was used to carry out the three factor and three level regression orthogonal test， and the regression models of qualified index， leakage index and multiple index were established respectively. The parameters of each influencing factor were optimized with the qualified index， leakage index and multiple index as the objectives. The optimal parameter combination was： the friction coefficient was 0.46， the seed height was 0.4 kg （in this paper， the total mass of seeds in the seeder was used to replace the seed height）， the rotate speed is 1.8 r/s （the linear speed is 9.36 km/s）， providing a maximum qualified index 98.03%， a minimum leakage index 0.40%， and the multiple index is 1.57%. Under these conditions， the qualified index is 92.86%， the leakage index is 6.43%， and the multiple index is 0.71% in the bench test， which has little error with the simulation results.