Search

The continuously growing size of deep neural networks (DNNs) has sparked a surge in research on model compression techniques. Among these techniques, multi-resolution model compression has emerged as a promising approach which can generate multiple DNN models with shared weights and different computational complexity (resolution) through a single training. However, in most existing multi-resolution compression methods, the model structures for different resolutions are either predefined or uniformly controlled. This can lead to performance degradation as they fail to implement systematic compression to achieve the optimal model for each resolution. In this paper, we propose to perform multi-resolution compression from a Bayesian perspective. We design a resolution-aware likelihood and a two-layer prior for the channel masks, which allow joint optimization of the shared weights and the model structure of each resolution. To solve the resulted Bayesian inference problem, we develop a low complexity partial update block variational Bayesian inference (PUB-VBI) algorithm. Furthermore, we extend our proposed method into the arbitrary resolution case by proposing an auxiliary neural network (NN) to learn the mapping from the input resolution to the corresponding channel masks. Simulation results show that our proposed method can outperform the baselines on various NN models and datasets.

Geothermal reservoir rocks have the characteristics of high stress, high hardness and high abrasiveness, which makes the drilling rate low and the drilling cost high, and hinders the application of geothermal energy. Improving drilling efficiency in hard rock strata is the prerequisite for large-scale application of geothermal energy. Impact drilling is considered to be the most effective drilling method in hard formations. PDC (Polycrystalline diamond Compact) bit compound impact drilling is one of the impact drilling methods, which has been widely concerned in recent years. In order to further improve the drilling efficiency of this method, the mechanism of PDC bit compound impact rock breaking is studied. Drucker–Prager criterion was used as rock constitutive relation, equivalent plastic strain was used as rock failure criterion, MSE (mechanical special energy) was used as evaluation method of rock-breaking efficiency, and a three-dimensional simulation model of full-size PDC bit for compound impact rock-breaking was built. The rock breaking law in PDC bit compound rock breaking process and the impact vector angle and impact duration on rock breaking efficiency are analyzed. The results show that the compound impact drilling can effectively alleviate the stick–slip vibration phenomenon, and improve drilling efficiency. In the compound impact mode, compressive stress failure and tensile stress failure occur at the same time, and the tensile stress failure is the main failure mode. The peak stress of the rock is 5.49% larger, and the residual plastic deformation is about 4.28% smaller than that of non-impact mode. With the increase of impact angle, the MSE decreases first and then increases a little. The optimal impact vector angle is in the range of 30°–50°. With the increase of the impact time, the MSE decreases in a wavy manner. In the range of 0.2 ms–1.0 ms, the optimal impact time is 0.8 ms–1.0 ms.

With the growth of neural network size, model compression has attracted increasing interest in recent research. As one of the most common techniques, pruning has been studied for a long time. By exploiting the structured sparsity of the neural network, existing methods can prune neurons instead of individual weights. However, in most existing pruning methods, surviving neurons are randomly connected in the neural network without any structure, and the non-zero weights within each neuron are also randomly distributed. Such irregular sparse structure can cause very high control overhead and irregular memory access for the hardware and even increase the neural network computational complexity. In this paper, we propose a three-layer hierarchical prior to promote a more regular sparse structure during pruning. The proposed three-layer hierarchical prior can achieve per-neuron weight-level structured sparsity and neuron-level structured sparsity. We derive an efficient Turbo-variational Bayesian inferencing (Turbo-VBI) algorithm to solve the resulting model compression problem with the proposed prior. The proposed Turbo-VBI algorithm has low complexity and can support more general priors than existing model compression algorithms. Simulation results show that our proposed algorithm can promote a more regular structure in the pruned neural networks while achieving even better performance in terms of compression rate and inferencing accuracy compared with the baselines.

Real-time temperature prediction is essential to circumvent thermal safety issues for lithium-ion batteries (LIB). However, its industrial applications are challenging due to operating temperature, voltage range, capacity degradation, and current rate (C-rate) variations. This study proposes a digital twin (DT) technology and long short-term memory (LSTM) neural network-based method for real-time temperature prediction and degradation pattern analysis. The DT model is established based on lumped thermal equivalent circuits to describe the dynamic thermal behavior of LIB and identify the parameters by calculations. Furthermore, the real-time temperature prediction framework considering voltage, current, and operating temperature is further designed following identifying results, which consists of indicators extraction, correlation analysis, LSTM neural network, and DT model four parts. In addition, the incremental capacity analysis (ICA) method is used to analyze LIB degradation patterns. The experimental results indicate that the primary degradation pattern of the experimental sample is loss of lithium inventory (LLI), and the loss of active material (LAM) appears after 600 cycles. Moreover, the maximum error and root mean square error (RMSE) of the temperature prediction framework is 0.31 °C and 0.17 °C under the constant current (CC) discharging condition, 0.85 °C and 0.47 °C under the dynamic discharging condition, respectively. The results demonstrate that the proposed real-time temperature prediction framework delivers acceptable accuracy for CC discharging and dynamic discharging conditions, which can complete the requirements of practical applications. This study dramatically reduces the response time of temperature prediction and guides optimizing battery thermal management systems (BTMS).

Gas drilling, as an important part of underbalanced drilling, can increase drilling speed. But in the process of conventional gas drilling, it tends to cause cutting settlement, borehole collapse, sticking and other safety hazards because gas circulation has to be interrupted. Therefore, this paper presents a continuous circulation sub which can be installed and removed easily. With this sub, gas circulation will not be interrupted when drilling tools are connected and removed. This sub is composed of body, main valve, bypass valve and side entry sub. The structure design of its key components (i.e. main and bypass valves) were fulfilled. Based on statics analysis on the sub body, its force situations under extension, torsion and internal pressure were simulated by using the ANSYS finite element analysis software. It is shown that its stress distribution trend is consistent with its elastic–plastic mechanics analysis results. Stress concentrates around the two round holes of the sub body, and the maximum deformation amount is still at the stage of elastic deformation. The analysis results are in line with the elastic–plastic mechanics analysis results, and the requirement of body strength is satisfied. This paper provides a new program to guarantee the drilling safety of extended-reach wells, underbalanced wells and narrow-density window wells.

For high-pressure low-permeability wells, wellbore temperature drops drastically in high-rate and multistage acid fracturing process. Under the combined action of the swelling of tubing string and the contraction of annular fluid between packers, annular pressure between packers undergoes violent transient change in staged acid jobs, thereby deteriorating loading on the tubing string and packers. Based on the principle of energy conservation and wellbore heat conduction, the transient prediction of two-dimensional (2D) wellbore temperature field under pumping injection condition was established by considering the effects of heat generated by friction and convection heat exchange. Moreover, the effects of wellbore temperature/pressure changes on the annular volume between packers were analyzed. Furthermore, in combination with the transient prediction model of wellbore temperature, PVT state equation of annular fluid, the calculation model of tubing string radial deformation and the transient seepage equation of the formation, the transient prediction model of annular pressure between packers in high-pressure low-permeability wells was established. Finally, by taking a high-pressure low-permeability well as an example, annular pressure between packers was calculated and the forces on the packers and tubing string were analyzed. According to the prediction results, the tubing string, which was regarded to be safe using conventional design method, exhibited an extremely high risk of failure after taking into account the decrease in annular pressure between packers. Therefore, the decrease in annular pressure should be fully considered in the design of tubing string for high-pressure low-permeability wells in multistage acid fracturing process. In combination with sensitivity analysis results, it can be concluded that formation permeability, injection rate and formation pressure all affected the change in annular pressure between packers.