Abstract:High-precision gravity measurement is an important observation method for inertial navigation, geodesy, space science, ocean exploration, and basic physics research. High-precision gravimeter has always been the goal pursued by scientific research. Atom interferometry gravimeter can provide absolute gravity acceleration of uGal magnitude. It is one of the most significant instruments to obtain high precision gravity information. The design and implementation of its compact laser system is of great significance for its practicality and commercialization. Firstly, the output requirements of atom interferometry gravimeter laser system are introduced, and then the related research and development of compact laser system at home and abroad are reviewed, and the implementation schemes and development status of two kinds of light source systems based on free space and optical fiber transmission are introduced respectively, with emphasis on its key technologies such as frequency stabilization, frequency hopping, power stabilization and so on. Finally, the compact laser system of atom interferometry gravimeter is summarized and prospected.

Abstract:Switch mode power supply (SMPS) is an important component of the electronic system, the fault state of SMPS has an adverse impact on the operation of the back-end components and the entire electronic system. Therefore, it is very necessary to identify the health state of SMPS. Under the environmental stresses, Multi-parameters of the components of SMPS will degrade. To effectively identify the state of SMPS, the paper presents the multi-parameter identification method based on the key features and Elman neural network. At first, the paper obtains the Wavelet Packet local energy features of the output. To improve the identification accuracy, the coefficient of variation are used to select the local energy features, the local energy features with lager coefficient of variation values were regarded as the key features. Finally, the relationship between the key features and parameters will be established based on Elman neural network. The results of the simulation and hardware experiments demonstrate that the proposed method can obtain the high identification accuracy and great practicability.

Abstract:Aiming at the problems of low efficiency of relay selection algorithm and security threat of potential eavesdropping nodes in multi-relay communication network. A joint relay and jammer selection strategy based on artificial neural network is proposed. Firstly, the decode-and-forward ( DF ) relay protocol is adopted to construct the multi-relay cooperative communication network with eavesdropper, and the closed form expression of the security outage probability is derived by combination with the cooperative jamming strategy. Then, the neural network is trained, and the channel state information (CSI) of the relevant nodes is taken as the input data to train the model to obtain the optimal model parameters. Finally, some data sets are used to verify the model, and the simulation results show that the accuracy of optimal nodes selection can reach more than 93%. Compared with the traditional selection scheme based on exhaustive search and support vector machine, the proposed scheme reduces the implementation complexity and computation time significantly, and effectively improve the security performance of the system.

Abstract:Sleep apnea syndrome, as a common sleep-related respiratory disorder, has gained a lot of attention. Due to its complicated diagnosis process and high price, it has attracted many researchers to explore fast and convenient detection methods based on singlechannel signals. The research proposes a multi-scale convolutional neural network method for rapid detection of sleep apnea based on ECG signals. Compared with the traditional single-scale convolutional neural network, the method can effectively combine the detailed and abstract information of the signal, and improve the feature representation ability of the convolutional neural network. By verifying on the Apnea-ECG database provided by PhysioNet, the proposed multi-scale convolutional neural network obtains an accuracy of 85. 2%, sensitivity of 83. 1% and specificity of 86. 5%. Compared with existing methods, the method further improves the performance of sleep apnea detection.

Abstract:Aiming at the cycle slip problem in the data processing of the Beidou navigation and positioning system (BDS), a method for detecting and repairing cycle slips based on lifting wavelet combined with NAR dynamic neural network is proposed. Firstly, the nondifference cycle slip test quantity is constructed, and the epoch of cycle slip is detected by the lifting wavelet method. Then, the NAR dynamic neural network method, the improved BP neural network method and the traditional polynomial fitting method are used to analyze and compare the effect of different methods on cycle slip repair. Experimental simulation results show that in cycle slip detection, the lifting wavelet method can effectively detect small cycle slips of more than 0. 2 weeks; in cycle slip repair, the NAR neural network improves the fit of the improved BP neural network by about 40%, and the prediction accuracy is about 50% higher than the improved BP neural network, and more than 10% higher than the traditional polynomial fitting method. It is more suitable for the detection and repair of small cycle slips, and further improves the positioning accuracy.

Abstract:Deep learning-based health indicator construction has become a new research and application hotspot in the field of machinery fault diagnostics. The performance of deep learning-based health indicators is largely depending on hand-craft feature extraction and selection. Moreover, the correlation of multi-channel sensor signals is not enough considered. In response to the above problems, a method for constructing health indicators based on multi-channel information fusion based on Deep Fusion Neural Network (DFNN) is designed. First, a multi-channel feature extractor (MFE) is proposed to extract bearing degradation features from the raw vibration signals. Then an adaptive feature selector (AFS) is designed to select useful features automatically. After MFE and AFS, we utilized a bidirectional long-short-term memory (BiLSTM) network to construct bearing health indicator. The proposed method is experimentally verified on the bearing life data set. The result shows that compared with some state-of-the art methods, the health indicator by DFNN is up to 98. 4%, and the monotonic indicator increases by 44%. Therefore, it is able to map the bearing degradation process effectively.

Abstract:Blood pressure is an important physiological parameter of the human body, which can reflect the pumping function of the heart, peripheral vascular resistance, and blood volume. Non-contact continuous measurement of blood pressure is of great significance in daily life and many applications. This paper obtains the relevant pulse wave signal from the facial video, and then extracts the characteristic parameters of the signal that are highly correlated with blood pressure, so as to use these parameters to establish a neural network model for blood pressure estimation and optimize it by genetic algorithm. Through verification, it is concluded that the genetic algorithm optimized BP neural network ( GA-BP) model estimation ability and fitting accuracy are significantly improved, and the estimation results meet the blood pressure measurement standards while realize the non-contact continuous estimation of blood pressure. The estimated accuracy rate of systolic blood pressure was 93. 1%, and the estimated accuracy rate of diastolic blood pressure was 96. 6%. Therefore, the establishment of GA-BP model by pulse wave characteristic parameters is an effective non-contact estimation method of blood pressure.

Abstract:Due to the vague definition of abnormal events and the scarcity of its own samples, the detection of video abnormal events has always been a challenging problem. Existing methods often separate the two steps of video feature extraction and anomaly detection model establishment, it leads to the method that cannot reach the optimum. This paper follows the idea of distance-based anomaly detection, and proposes a one-class neural network method for video anomaly detection. This method combines the layer-by-layer data representation ability of the autoencoder and the one-class classification ability. The features of the hidden layer are constructed for the specific task of anomaly detection, thereby obtaining a hyperplane to separate all normal samples from abnormal samples. The experimental results on two benchmark data sets show that the proposed method achieves 94. 9% frame-level AUC and 94. 5% frame-level AUC on the PED subset and PED2 subset, respectively, and achieves 80 correct event detections on the Subway dataset, confirming the wide applicability of the method in industrial and urban environments.

Abstract:Aiming at the shortcomings of the traditional translational scanning detection system, a three-dimensional detection and reconstruction system based on rotating scanning line structured light and the corresponding system parameter calibration method are proposed, and a point cloud data acquisition model is established. The measured object makes a relative movement with linear structured light by rotation, and a two-dimensional image of the external surface of the measured object is obtained. The system calibration obtains the conversion relationship between the coordinates of the image and the coordinates of the world, and obtains the information of threedimensional coordinates and numerical model of the measured object. It can be seen from the experiment that the calibration accuracy of the camera is 0. 2 mm, and the accuracy of the principle prototype for object measurement is 0. 1 mm. Experiments prove that the system has high detection accuracy and feasibility.

Abstract:Aiming at the problems in the current FPGA welding point failure assessment methods, such as the inability to provide accurate information, lack of sample data and low timeliness, combined with genetic algorithm (GA), an improved FPGA welding point failure assessment method based on the least square support vector machine (GA-LS-SVM) was proposed. Establish the SJ BIST test model, select the appropriate small external capacitor, simulate the welding spot resistance value by changing the variable resistor size at different operating frequencies, obtain the fault data based on the voltage change of small capacitor, and establish the three-dimensional data graph of the duration of capacitor low level, capacitor test working frequency and welding point resistance value; Finally using genetic algorithm to optimize the least squares support vector machine (SVM) to state evaluation of the obtained data, according to the three-dimensional data graph, there is a significant difference in the duration of low-level between healthy FPGA solder joints and broken FPGA solder joints. The simulation results show that the proposed GA-LA-SVM method has an overall accuracy rate of 97. 2%, which is 17. 9%, 13% and 7. 2% higher than BPNN, standard SVM and LS-SVM methods.

Abstract:For the low implementation complexity requirement of the low-density parity-check ( LDPC) encoders in consultative committee for space data systems (CCSDS) standard, an implementation architecture of multi-channel parallel encoder is proposed for LDPC codes with different code lengths and code rates. The matrix information can be shared among all parallel computing units to improve resource utilization by repeatedly utilizing the storage unit in the encoder. Furthermore, the single-channel and multi-channel encoders with code rates of 1 / 2, 2 / 3 and 4 / 5 are verified and tested on the field programmable gate array (FPGA) hardware platform. The test results show that the throughput of the encoders adopting the multi-channel parallel coding scheme is higher than that of the single-channel encoders and achieves more than 1Gbps. The resources of the look-up table for the multi-channel encoders are reduced by 40%, 44% and 46%, respectively, compared with the single-channel encoders with multiple groups that achieves approximately the same throughput. By making full use of the storage resources in FPGA, this architecture can reduce the complexity of hardware implementation effectively.

Abstract:In recent years, quality-related fault diagnosis has attracted much attention from academic circles. Several fault diagnosis algorithms based on post-processing have been developed. However, further studies have found that these post-processing methods will gradually lose their function when the magnitude of quality-unrelated faults increases. In addition, post-processing algorithms will generate a lot of calculations in practice. In order to further solve the drawbacks of the above methods, this paper adopts a structure of preprocessing, modeling, and postprocessing. And proposed a modified orthogonal projection to latent structures (MOPLS). Compared with the previous algorithm, this method is more practical for quality-related faults. At the same time, the number of latent variables required by the model is reduced. Therefore, its computational complexity is lower than the previous algorithm. Numerical examples and Tennessee-Eastman (TE) process are used to verify the effectiveness of the method.

Abstract:In order to solve the problems of low tracking accuracy, large oscillation and poor real-time performance of ball and plate system, a sliding mode control scheme of combinatorial reaching law based on genetic algorithm is proposed. Firstly, the sliding mode surface designed by Ackermann formula is used to express the discontinuous hyperplane; secondly, the exponential reaching law is used to reduce the buffeting amplitude in the early stage of the sliding mode, and the variable speed reaching law is used to converge the buffeting in the later stage. Finally, the genetic algorithm is used to select the parameters. the experimental results show that the initial output amplitude of the combined reaching law is reduced by 2. 2 compared with the variable speed reaching law, and the convergence time after genetic algorithm optimization is reduced by 0. 4 s. The stability of the controller is proved by Lyapunov theory, and the simulation results show that the control strategy has good dynamic quality and steady-state performance, and meets the trajectory tracking requirements of the ball and plate system.

Abstract:Magnetic flux leakage internal detection technology is one of the means for detecting defects in long-distance oil and gas pipelines. The identification of geometric features of defects is of great significance to the safety operation evaluation of pipeline. Based on the two-dimensional magnetic dipole model, a three-dimensional analytical model of the spatial distribution of the magnetic flux leakage field of the inner wall defects of the pipeline is established, and the variation law of the magnetic flux leakage field generated by the magnetic charge when the magnetization direction is perpendicular to the defect is studied. Based on the analytical model of the inner wall, a demagnetization factor is introduced to compensate the model. The three-dimensional analytical model of the defect leakage magnetic field on the outer wall of the pipeline is established, and the distribution characteristics of the defect magnetic flux leakage field on the outer wall of the pipeline are obtained. An experimental platform for magnetic flux leakage detection is built to verify the effectiveness of the model. The results show that the magnetic flux leakage field of the outer wall has a certain shielding effect, and the experimental results are in good agreement with the theoretical analysis. The model can effectively describe the spatial distribution characteristics of the magnetic flux leakage field on the inner and outer walls of the pipeline, which has certain engineering guiding significance for defect identification and quantitative evaluation.

Abstract:Time-to-digital converter (TDC) is a time interval measurement circuit, widely used in time-of-flight (ToF) measurement, frequency measurement and other fields. Aiming at the problem that traditional TDC was constrained by the mutual restriction of resolution and measurement range, a novel TDC with two-step quantization that gave consideration to both resolution and measurement range was designed in SMIC 55 nm CMOS process. The first stage used ring structure for coarse quantization, which improved the measurement range. The second stage used a delay locked loop (DLL) to generate the control voltage that voltage-controlled delay cells needed and improved the resolution by scaling the load capacitors of the delay cells. This paper proposed a simple algorithm of time residue generation that transmitted the time interval cannot be quantified in first stage to second stage. The delay cell structure in first stage was designed in order to eliminate the delay mismatch that multiplexer caused when signal circling. The simulation results showed that the proposed TDC could realize the resolution of 4. 8 ps, the measurement range of 1. 26 μs. The measured maximum differential non-linearity (DNL) is 0. 6 LSB. The measured maximum integral non-linearity (INL) is 1. 8 LSB.

Abstract:Combined with Maxwell’ s equations, a mathematical model of light propagation in electro-optic crystals was established; physical parameter models of lithium niobate crystal and bismuth germanate crystal were established through finite element simulation, and a universal three-dimensional wave field simulation method of incident light was proposed. This method combines the electric field distribution inside the crystal with the coupled wave theory of the electro-optic effect, and calculates the Pockels effect of the two crystals under transverse modulation. The difference between the results and the theory is within 10 -5 orders of magnitude; at the same time, the simulation results show that the Pockels effect of lithium niobate crystals is more obvious. Finally, 0 ~ 1 000 V DC voltage experiments are carried out on two electro-optic crystals of different materials. The experiment shows that the measurement accuracy of the Pockels effect of the lithium niobate crystal is better than ±2. 9%. This method provides new research ideas and theoretical references for the selection and performance evaluation of electro-optic crystals for optical voltage sensors.

Abstract:In view of the non-stationary and random characteristics of the short-term traffic flow sequence, in order to improve the shortterm traffic flow prediction accuracy and model training speed, this paper proposes a combined prediction model based on adaptive variational modal decomposition (VMD) and bi-directional long-term memory network (BiLSTM) combined with attention mechanism. Firstly, the spatial and temporal traffic flow sequence is decomposed by the adaptive VMD method to a series of modal components with limited bandwidth, which can refine the traffic flow information, reduce non-stationarity, and improve the accuracy of modeling. Secondly, the spatio-temporal correlation in the short-time traffic flow sequence after decomposition is mined by BiLSTM combined with attention mechanism to reveal its spatio-temporal variation rules, which further improves the modeling accuracy. In addition, in order to accelerate the training convergence speed of the prediction network, the network weight optimization is carried out by the improved Adam algorithm. Finally, the predicted value of each modal component is superimposed as the predicted value of the final traffic flow prediction value. The experimental results show that the prediction performance of the model using modal decomposition is obviously better than that of the model without modal decomposition, and the RMSE of the self-adaptive VMD-Attention-BiLSTM prediction model is reduced by 47. 1% compared with that of the EEMD-attention-BiLSTM prediction model. The combined prediction model improves the prediction accuracy and can quickly predict the traffic flow time series.

Abstract:Aiming at the problems of voltage overshoot and current surge caused by parallel active power filter ( SAPF) grid connection under traditional control mode. Based on the protection of switching devices and line overcurrent protection from excessive current, the article analyzes the causes of voltage overshoot and current impact by establishing a mathematical model of SAPF DC side. After the above work, the transition process and first-order auto disturbance rejection control are introduced on the basis of the traditional series current limiting resistor start strategy, and a first-order auto disturbance rejection SAPF grid-connected start strategy with transition process is designed, which verifies the stability of the first-order active disturbance rejection control. The proposed startup strategy was verified in the simulation software, and compared with the traditional proportional integral (PI) control method in the simulation, the validity and feasibility of the proposed startup strategy were verified. Finally, through the simulation analysis, this startup strategy can suppress the inrush current generated during the grid-connected startup process of the system within 100 A. Compared with the traditional control strategy of 500 A, the inrush current greatly suppresses the damage to the grid; The voltage overshoot during grid connected startup is 0, and the SAPF has no impact start.

Abstract:In order to improve the image quality of uncooled infrared thermal imager, and meet the needs of viewing and locking in low contrast and dim-area, a super-resolution reconstruction model of infrared image based on multi-scale dense residual network is proposed in this paper. The basic framework of the model is to reconstruct high-resolution image by cascading multiple residual features. Firstly, a multi-scale cross-channel fusion module is proposed. By fusing the branch results of different receptive fields, it not only fuses the complementary information of different receptive fields, but also helps to improve the gradient convergence and feature transmission. Then, multiple cross-fusion modules are cascaded and optimized by residual feature learning to learn high-resolution detail information. The simulation results show that the super-resolution model proposed in this paper can achieve better super-resolution reconstruction effect, and has better performance in weak structure maintenance and point target maintenance. Our proposed model has achieved highquality super-resolution reconstruction on the embedded deep learning platform of Hisilicon, and has high engineering application value.

Abstract:The line laser sensor is a new type of sensor that has been more widely used in recent years. It has the advantages of noncontact, high accuracy and fast speed, which can acquire high-resolution point cloud data in a short time. However, the traditional point cloud registration algorithm has the problems of low registration accuracy and long registration time when processing the line laser point cloud, which make it difficult to meet the requirements of production time and accuracy in practical application scenarios. In this paper, we propose a line laser point cloud registration method that combines contour features to extract the contour features of the line laser and utilize them as the key points for point cloud registration iteratively, and experiments are conducted to compare the traditional iterative closest point. The experimental results demonstrate that our method has great potential for application because of its high accuracy and short registration time.

Abstract:In order to overcome the shortcomings as spectral distortion and poor target content saliency of the fused image by ignoring spectral features in current visible and infrared image fusion methods, the infrared and visible image fusion algorithm based on nonsubsampled contourlet transform coupled with feature selection mechanism is proposed in this paper. Firstly, the visible and infrared images are calculated by NSCT to separate it into different image coefficients. Then, the information entropy function is used to measure the richness of the image information content for forming the fusion coefficient of low-frequency coefficient, which can obtain the fusion low-frequency coefficient with rich information such as infrared target. The neighborhood information of pixels was used to measure the definition feature of image, and the mean function was introduced to measure the spectral feature of image. Through the definition feature and the spectral feature of image, the feature selection mechanism was constructed to select the ideal high-frequency coefficient fusion function from the image, and obtain the fused high-frequency coefficient that takes into account both the detailed characteristics and the spectral characteristics. Finally, the experimental results show that compared with the existing fusion algorithm, the proposed algorithm has better spectral characteristics, significant target content and better fusion performance.

Abstract:At the aspect of time of flight (TOF) based laser ranging, the interference pulse is superimposed on the rising edge of signal pulse, so the classic double-threshold processing algorithm cannot correct the identification error of pulse front time. In this paper, a dynamic multi-threshold fitting error correction algorithm is proposed. By switching three thresholds dynamically, the best fitting method is found to avoid the wrong fitting correction result caused by the interference pulse. In addition, a new concept of threshold time difference correlation curve is proposed, which adjusts the threshold dynamically according to the deviation degree between correction point and correlation curve. The experimental results show that the proposed dynamic multi-threshold error correction method can obtain ideal correction results, allows incorrect data to be filtered out, so as to improve the ranging accuracy of long-range targets.

Abstract:Aiming at the problem of clock frequency drift in IEEE1588 clock synchronization process, a novel clock synchronization algorithm based on sliding mode control is proposed. Firstly, the state space model of the system is established according to the recurrence relationship between master-slave clock offset and drift. Then, the sliding mode control is used to reduce the clock skew and clock drift. Finally, the frequency jitter and random error in the experimental process are optimized by sliding average filter. The results show that the clock synchronization algorithm based on sliding mode control can effectively restrain the linear growth of clock skew caused by clock drift, and control the clock skew below 1 μs, thus, realizing sub microsecond network time synchronization. Compared with the traditional IEEE1588 protocol synchronization method, the proposed method provides higher synchronization accuracy.

Abstract:In order to improve the problems of high-frequency chattering, phase delay, and low-speed failure in traditional sliding mode observers (SMO), this paper proposes an improved SMO control strategy. This method introduces an amplification factor before the lowpass filter to amplify the back-EMF signal for easy extraction and estimation; adopts the hyperbolic tangent function as the sign function to reduce the high-frequency chattering and signal interference of the system; adopts an adaptive filter to improve detection Accuracy, reduce phase delay. Through MATLAB/ Simulink simulation analysis, the two strategies of improved SMO and traditional SMO are compared. The results show that the improved SMO reduces the phase delay time by 97. 5% compared with the traditional SMO, and it can significantly reduce chattering phenomenon, and accurately estimate the motor below the speed of 20 r/ min. , to achieve good estimation performance at full speed, indicating that the improved SMO strategy has better estimation accuracy and adaptability.

Abstract:In order to improve the accuracy of rolling bearing performance degradation assessment for one class support vector machine (OCSVM), an ensemble empirical mode decomposition method based on adaptive white noise was proposed. A performance degradation evaluation method combining CEEMDAN, particle swarm optimization ( PSO) and One class SVM. Firstly, CEEMDAN was used to expand the collected vibration signal calculation into intrinsic mode functions ( IMFs), and typical characteristic signals were obtained according to the IMFs energy density. Secondly, the parameters ν of OCSVM and radial basis kernel function g are optimized by particle swarm optimization to enhance the learning ability and generalization ability of OCSVM. Finally, 3σ was used to set the adaptive threshold, determine the early failure threshold of the bearing and verify the accuracy of the evaluation results by using the CEEMDAN and Hilbert envelope demodulation method. The validity of the proposed model was verified by bearing experimental life data from the University of Cincinnati. The results show that the PSO algorithm optimized OCSVM model can accurately monitor the bearing running life state. Compared with the support vector data description ( SVDD) and parameter optional OCSVM model, the performance degradation assessment model of this method is more effective and superior.

Abstract:The conduction phase inductance of switched reluctance motor will change with the change of the conduction phase current under the case of magnetic circuit saturation, resulting in a low rotor position estimation accuracy. Aiming at this problem, a position sensorless control method of switched reluctance motor considering magnetic circuit saturation is proposed. Firstly, the relationship formula of the phase inductance function is established, and then the relationship between the phase current and the position angle of the extreme positioning point of the phase inductance is analyzed. Finally, the realization process of estimating the corresponding rotor position angle from the two adjacent extreme positioning points is described. Finally, the related simulation and experiment were carried out with a three-phase 6 / 4 structure motor. The results show the feasibility of the above method.

Abstract:The detection of the leakage current of the zinc oxide arrester is achieved by extracting the fundamental and harmonic parameters of the leakage current through harmonic analysis, and the operation of the device in the power grid can be judged according to the resistive component of the harmonic current. In order to solve the problem of digital signal processing performance degradation caused by data truncation during harmonic analysis by fast Fourier transform (FFT). All-phase FFT analysis method that considers all possible data truncation conditions of a sample point is selected, and then all-phase time-shift phase difference correction algorithm is selected based on this analysis method, and use the window spectrum function to derive a correction formula. Compared with the ratio correction algorithm based on FFT, the algorithm which is selected has improved the accuracy of harmonic analysis by 4 to 5 orders of magnitude in the case of no noise and 1 to 2 orders of magnitude in the case of noisy (50 dB), and the frequency estimation deviation for relatively small signals does not exceed 0. 6 Hz, and the phase estimation deviation does not exceed 0. 5°. This algorithm has phase invariance and better ability to suppress spectrum leakage and realizes the improvement of detection accuracy and the estimation of small signal harmonic parameters, which is verified in the leakage current detection system.

Abstract:Chinese railway lines are widely distributed in the windy and snowy areas. The study of the deposition characteristics of snow crystal on the surface of the cantilever insulator is of great significance in preventing snow flash in the winter and improving the reliability of power supply. In this paper, the finite element simulation software was used to simulate amount of snow crystal deposition of railway insulators under wind speed. The velocity and static pressure distribution of flow field on insulator surface are calculated to analyze the influence of wind speed, wind direction and snow crystal type on the surface deposition of insulators. The simulation results show that the snow crystal type has little effect on the insulator amount of snow crystal deposition under the same wind speed and direction. Solid dendritic snow crystal deposition is the largest. The amount of snow crystal deposition of insulator was installed horizontally increased with the increase of wind speed and the amount of snow crystal deposition of insulator was installed obliquely increased first and then decrease with the increase of wind speed, When the wind angle is negative, the installation method has a serious impact on the snow crystal deposition. The relationship between amount of snow crystal deposition on the surface of insulator and wind direction presents a “V”-shaped distribution when the wind speed was 10 m/ s and wind direction was within ±45°. The snow crystal deposition distribution on the surface of shed increases with the increase of the wind direction, and the snow crystal deposition on the windward side of shed is larger than that on the leeward surface of the shed.