1146 results about "Multiple stages" patented technology

Apparatus and methods are disclosed for implementing a copilot as a network of microservices including specialized large language models (LLMs) or other trained machine learning (ML) tools. This architecture supports flexible, customizable, or dynamically determinable dataflow. Compared to much larger competing LLMs, comparable or superior performance is achieved for certain tasks, while significantly reducing computation time and hardware requirements, even to a single compute node with a single GPU. An expansion microservice converts client input tokens into associated tokens to diversify targeted tasks presented to a core microservice. An ML tool in the expansion microservice is pretrained in multiple stages for varying tasks using varying general and target-specific corpora. Synthesized training data derived from a knowledge graph can also be used. The expansion ML tool is subsequently fine-tuned in one or multiple stages. Variations and additional techniques are disclosed.

The invention relates to the technical field of hydrogen production, in particular to a bipolar plate runner structure-mass transfer efficiency parameter simulation optimization method. Comprising the steps that a flow channel structure is designed, specifically, a bipolar plate flow channel is constructed through a fractal tree-shaped network topological structure, the flow channel is composed of multiple stages of branch channels, the section size of each stage of branch channel is decreased progressively according to a self-similarity proportion, and a three-dimensional flow channel network with fractal dimensions is formed; multi-physics field coupling modeling: establishing a multi-physics field coupling model including fluid flow, electrochemical reaction and heat and mass transfer based on computational fluid mechanics and a finite element method, wherein the model considers a turbulence effect in a flow channel, interface impedance of a porous medium diffusion layer and reaction kinetic parameters at the same time; parameter sensitivity analysis: screening key parameters influencing mass transfer efficiency through orthogonal test design, including flow channel fractal dimension, branch angle, porosity and surface wettability parameters, and constructing a parameter response curved surface; the flow channel structure of the bipolar plate can be accurately optimized, and the mass transfer efficiency is effectively improved.

Provided in embodiments of the present disclosure are a method for scheduling workloads of a data center in an electricity market environment, and a device. In the embodiments of the present disclosure, a multi-stage random optimization solution that allows a data center to participate in a multi-level electricity market is provided. The types and data dependency relationship of workloads are taken into consideration, and a load scheduling model, in a time dimension, for the workloads is constructed; on the basis of electricity source information of the data center, a multi-stage electricity cost optimization model is constructed, so as to comprehensively consider the exogenic uncertainty of an electricity price and carbon emissions and the endogenic uncertainty of the workloads; on the basis of the multi-stage electricity cost optimization model, multi-stage optimization solving is performed on the load scheduling model, so as to obtain a load scheduling scheme and an electricity purchase strategy. Therefore, a more accurate and more reliable estimation is made concerning the electricity flexibility of the data center, such that the electricity cost of the data center can be reduced, support is provided for the reduction of carbon emissions, and the consumption of renewable energy sources is encouraged, thereby achieving the energy use flexibility of the data center in the multi-level electricity market environment.

The invention discloses a method for dynamically adjusting the frequency of a resonant cavity of a cyclotron and a tuning circuit, which are applied to the field of cyclotrons and solve the problem that when the frequency of the resonant cavity of the cyclotron is adjusted by judging the magnitude of reflection power, due to the fact that the reflection power can change in multiple stages in a single macro pulse period, the frequency of the resonant cavity of the cyclotron is adjusted. And an accurate numerical value is difficult to obtain through a simple amplitude sampling circuit. According to the invention, through determining the stable resonance time period of the resonant cavity, the phase difference between the incident power signal and the reflection power signal is sampled in the corresponding stable resonance time period of the resonant cavity, and the change value of the phase difference between the incident power signal and the reflection power signal is dynamically compared. According to the conversion relation between the phase difference change value and the frequency change value, the resonant cavity frequency change value is obtained through conversion, then the resonant cavity tuning device is adjusted to enable the resonant cavity frequency to be consistent with the frequency of the incident power signal all the time, and the dynamic tuning method of the resonant cavity is achieved.

The invention relates to a method for improving the calculation speed of a model based on a mercuric chloride AI processor. The method comprises the following steps: deploying the mercuric chloride AI processor and a deep learning model in a server, and carrying out adaptation and optimization on the mercuric chloride AI processor; before the data in the first buffer area is read, predicting and preloading the data to be processed, and loading the data from the global memory to the second buffer area in advance; a double-buffer mechanism is arranged in the mercuration AI processor, interrupt and event trigger points are set, the operation of loading data to a next buffer area is immediately started when a specified calculation stage is finished, and the time sequence of data circulation is accurately controlled; and a control parameter is automatically adjusted based on monitoring data of the real-time monitoring module, the calculation process is decomposed into a plurality of stages, different buffer areas are allocated for each stage, and access to the shared memory is optimized by setting a specified cache replacement strategy and the size of a cache line. According to the process, the calculation speed of the model in the deep learning field is improved, the resource utilization rate is improved, and the defect of manual adjustment and optimization is overcome.

The invention provides a pole piece and a manufacturing method thereof, a rolling system and a secondary battery, and the manufacturing method of the pole piece comprises the following steps: coating slurry on a current collector and drying to obtain a dried pole piece; the difference between the thickness of the dried pole piece and the preset thickness after rolling is used as the total rolling reduction; the dried pole piece is subjected to continuous multi-stage rolling, adjacent two-stage rolling is carried out, the percentage of the rolling reduction of the previous-stage rolling in the total rolling reduction is larger than or equal to the percentage of the rolling reduction of the next-stage rolling in the total rolling reduction, and the sum of the rolling reduction of all stages of rolling is equal to the total rolling reduction; in the multi-stage rolling, the rolling reduction of each stage of rolling is gradually reduced on the whole; and the multi-stage rolling is cold pressing. According to the manufacturing method of the pole piece, the multi-stage rolling mode is adopted, the rolling reduction decreasing and roller gap adjusting mode is adopted, the diameter, roughness and hardness of the roller body are combined to adapt to different rolling reduction, and finally the consistency of stress release and the pole piece is achieved.

The invention relates to a conflict-free AGV scheduling method for punctualized material distribution in a workshop, and the method comprises the following steps: obtaining a material distribution task, and dividing a decision period into a plurality of stages; an independent scheduling algorithm is used in each stage for AGV scheduling, tasks and departure time distributed by each AGV in each trip are optimized, a conflict-free optimal path is planned, an AGV scheduling scheme for minimizing time window penalty and distribution cost is realized, the scheduling algorithm comprises task allocation and path planning, in the task allocation, the path planning is performed, and in the path planning, the time window penalty and the distribution cost are reduced. According to the method, travel division is carried out based on departure time estimation, AGV assignment of each travel is carried out based on a heuristic rule, initial path planning is carried out by adopting an A * algorithm according to a task allocation result in path planning, conflict detection is carried out based on a path time window, and conflict resolution is carried out in combination with waiting and local re-planning. Compared with the prior art, cooperation of task allocation and path planning under the punctualization distribution requirement is achieved, and the calculation efficiency and the resource utilization rate can be improved.

The invention relates to a BIM-based deep foundation pit construction dynamic monitoring method and system, and specifically, the method comprises the steps: dividing a BIM design model into a plurality of stages according to a construction stage, and obtaining deep foundation pit site point cloud data and a current construction stage; constructing a point cloud picture based on the affiliation consistency probability and boundary conformity of the points in the point cloud in the BIM design model at the current construction stage, and segmenting the point cloud picture by adopting minimum segmentation with hierarchical constraint to obtain a plurality of point cloud clusters; calculating the internal geometrical morphology of the point cloud cluster based on the characteristic value of the covariance matrix of the point cloud cluster, and associating the point cloud cluster with the construction member in the BIM design model at the current construction stage by using the position relationship between the geometrical morphology and the point cloud cluster; and performing three-dimensional reconstruction on the point cloud cluster by using construction components associated with the point cloud cluster, and updating and visually displaying the current construction progress state of each component in the BIM design model.

The invention relates to a heterogeneous cluster-oriented resource allocation method and device and a storage medium, and the method comprises the steps: executing a plurality of iterations of an initial model on a single node of each heterogeneous cluster, and obtaining the performance characteristics of the model; generating a plurality of parallel strategy combinations comprising the first parallelism degree and the second parallelism degree; generating a plurality of assembly line parallel combinations based on the first degree of parallelism, calculating the load imbalance rate of each assembly line parallel combination, and screening out N combinations with the minimum load imbalance rate from the assembly line parallel combinations; traversing the second parallelism degree and the micro-processing batch in each stage corresponding to each parallel combination to generate a plurality of stage strategy combinations; and calculating video memory occupation and execution time under each stage strategy combination, screening out K stage strategy combinations of which the video memories are smaller than or equal to a video memory threshold value and the execution time is shortest, and carrying out resource allocation on the heterogeneous cluster based on the stage strategy combinations. According to the invention, the problems of low training efficiency and low resource utilization rate are solved.

The invention discloses a deformable wing with the characteristics of large unfolding-folding ratio and high bearing capacity. The deformable wing comprises a bearing structure, a driving mechanism, a wing surface dimensional structure and an accessory structure, the bearing structure comprises a plurality of groups of telescopic mechanisms, wing ribs and corner boxes; the accessory structure comprises a mounting plate, and the mounting plate is connected with the machine body; each group of telescopic mechanisms comprises a plurality of stages of telescopic rods which are arranged along the chord length direction and are fixed on the mounting plate; the wing ribs are arranged in the wingspan direction and are connected with the multi-stage telescopic rods through the angle boxes to form a wing framework; the driving mechanism penetrates through the hollow parts of the wing ribs, and two ends of the driving mechanism are respectively connected with the mounting plate and the wing rib on the outermost side of the wing; the wing skeleton is externally coated with the wing surface dimensional structure to form a wing shape.

The invention relates to the field of grinding machinery, and discloses a cylinder body casting grinding device and a grinding process thereof.The device comprises a frame body and a controller, a three-jaw chuck is arranged on the frame body, and a to-be-ground cylinder body is clamped through the three-jaw chuck; a grinding mechanism is further arranged on one side of the frame body and comprises a hydraulic telescopic cylinder and a driving part, the hydraulic telescopic cylinder is fixedly installed on the frame body, the driving part is fixedly installed at the end of the hydraulic telescopic cylinder, and the output end of the driving part is in transmission connection with a grinding shaft; a grinding shell is fixedly arranged at the end of the grinding shaft, multiple stages of grinding wheels are arranged in the grinding shell, an alternating assembly is further arranged in the grinding shell, and the alternating assembly is used for pushing the different stages of grinding wheels to sequentially stretch out of the grinding shell to conduct the grinding action of the cylinder body. By means of the mechanism, real-time replacement of different levels of grinding wheels in the working state in the cylinder body is achieved, and the grinding efficiency is higher.

The invention discloses an efficient hybrid parallel training method for a multi-modal model, and belongs to the technical field of computers. The method comprises the steps that firstly, a multi-modal model architecture is divided into a plurality of independent modal sub-modules according to the multi-modal model architecture, and each sub-module processes one kind of modal data; secondly, formulating a computing power distribution scheme by statically analyzing the computing load of each sub-module; calculating time and parameter quantity of each layer of a modal sub-module are analyzed, then the layers are divided into a plurality of stages by using a partition algorithm, and asynchronous pipeline parallel and data parallel are combined to accelerate training; according to the method, the hardware resource utilization rate is remarkably improved through the calculation power distribution and hybrid parallel strategy for the modal calculation load, and the acceleration of the multi-modal model training process is realized.

The invention discloses an anti-explosion pressure relief ball valve which comprises a pressure monitoring system and a control execution system. The pressure monitoring system is used for forming a pressure distribution model; the control execution system comprises a signal processing module used for calculating the pressure change rate; the abnormal pressure identification module is used for judging whether an abnormal pressure event exists or not according to the judgment condition; the multi-stage pressure relief control module is used for controlling the opening number of the micro-channel systems and controlling the pressure relief amount according to the relation between the current pressure and a preset threshold value; the stepped pressure relief structure comprises multiple stages of pressure relief cavities, so that the pressure relief process conforms to a pressure change formula; and the dynamic feedback control module is used for dynamically adjusting the opening amount of the micro-channel by adopting a PID control algorithm. The system has the following beneficial effects that through pressure monitoring, intelligent judgment and the multi-stage controlled pressure relief module, early recognition, accurate intervention and safe disposal of abnormal pressure events are achieved, and safety accidents such as pipeline burst and flange leakage are effectively prevented.

The invention discloses a lifting stage structure control system and method, and relates to the technical field of stage control, and the lifting stage structure control system and method comprises the steps: S1, collecting the multi-source state data of stage equipment in real time, carrying out the data preprocessing, and constructing a multi-source state database; s2, constructing a motion state space model, predicting a time sequence motion sequence at a future moment, judging whether stage equipment is overlapped or not, predicting a collision risk, and taking risk avoiding measures; s3, the obstacle avoidance priority of the stage equipment is judged, a stage equipment movement scheme is generated, and a control instruction is issued; s4, motion operation is executed, the health degree of motion execution of the stage equipment is evaluated in real time, and graded early warning measures are taken; s5, displaying the stage situation and risk based on the collision risk and health degree evaluation result, and generating an analysis report; the problem that potential collision risks cannot be avoided in time due to dynamic path intersection caused by cooperative operation of multiple pieces of stage equipment in a complex stage scene is solved.

The invention relates to an underground engineering construction design technology, discloses a method for analyzing the influence of underground excavation under-crossing on an existing subway station, and improves the efficiency and precision of analyzing the influence of underground excavation under-crossing construction on the existing subway station. The method comprises the steps that by defining a structure and geometrical shapes and material attributes of a stratum, boundary conditions and external loads are set, grid division is carried out, and a finite element three-dimensional basic calculation model is constructed; according to theories of porous media and the like, a multi-physical field coupling model of a seepage-deformation-temperature full coupling effect is constructed; the rigidity of the underground excavation vertical pile foundation is calculated, and the rigidity of the underground excavation vertical pile foundation is corrected based on the influence factors of the soil squeezing effect and the negative friction resistance; subsurface excavation construction is subdivided into a plurality of stages, structural stress, deformation and stratum displacement are calculated through an incremental method, and the timeliness of a supporting structure and the influence of different excavation modes on stratum disturbance are analyzed; and obtaining a structural response through finite element iteration solution, and judging the influence of underground excavation under-crossing on the existing subway station.

In certain embodiments, model unlearning verifiability may be facilitated. In some embodiments, in connection with a request for unlearning of a dataset for a machine learning model, an unlearning process may be performed on the machine learning model, where the unlearning process involves multiple stages, for the unlearning of the dataset, at which model instances of the machine learning model are respectively generated. One or more proofs or hash-based values corresponding to the model instances may be generated, such as first and second hash-based values corresponding to first and second model instances of the machine learning model generated at first and second stages of the unlearning process, respectively (e.g., where the second stage depends on the first stage). The proofs or hash-based values may then be stored at one or more database locations of a database to enable verification of the performance of the unlearning process.

The invention relates to a filling mining method stope structure parameter optimization method and system based on numerical simulation, and the method comprises the steps: obtaining mine engineering geological conditions, building a mine entity model, and determining a stope structure parameter scheme; establishing a stope model according to the determined stope structure parameters; combining the mine entity model and the stope model, re-grouping the combined model, and assigning values to parameters in the combined model by setting mechanical parameters extracted from initial conditions and mine engineering geological conditions; and carrying out excavation and filling simulation on the combined model, respectively determining the stability indexes of the stope in the states of the original rock, the goaf and the filling body, and screening out an optimal stope structure parameter scheme. The problem that the resource development efficiency is affected due to the fact that an existing stope structure parameter optimization scheme is poor in analysis efficiency and accuracy in mine stability analysis under the multi-stage, multi-time-sequence and multi-scale conditions is solved.

The invention provides a three-dimensional point cloud segmentation method and system based on bidirectional fusion of a point cloud and an aerial view, and belongs to the technical field of three-dimensional scene environment perception, and the method comprises the steps: obtaining to-be-segmented original point cloud data; processing the obtained original point cloud data by using a pre-trained segmentation model to obtain a segmentation result; wherein the segmentation model comprises an encoder module, a backbone network and a fusion segmentation head. According to the method, a bidirectional fusion mechanism of the point cloud and the aerial view is provided, through interactive fusion of the point cloud and the BEV features in multiple stages, the expression ability of the model for local and global features is improved, the problems of fuzzy semantic boundary, difficulty in small target recognition and the like are remarkably improved, and the segmentation precision is obviously superior to that of an existing BEV method. Compared with a voxel method and a multi-representation fusion method, the method has the advantages that the weight of a calculation graph is kept light in structural design, compared with the voxel method and the multi-representation fusion method, the method has remarkable advantages in the aspects of parameter quantity and reasoning speed, precision and real-time performance are better considered, and the method is suitable for scenes with extremely high efficiency requirements such as automatic driving.

The invention discloses a plane matrix type high folding and unfolding bionic antenna truss structure with a negative Poisson's ratio effect, and relates to the technical field of space folding and unfolding mechanisms, the plane matrix type high folding and unfolding bionic antenna truss structure comprises a plurality of cell elements and a plurality of cell element connecting rods, transverse cell elements are connected through the cell element connecting rods, and longitudinal cell elements are fixedly connected through adjacent cell element cross beams; the cell element connecting rods are hinged to the cell elements on the two sides respectively, and the cell elements are three-level positioning inwards-concave hexagonal cell elements; the folding and unfolding antenna truss structure is in a folded state during launching, transportation and silent latency and is in an unfolded working state during on-orbit task execution, the high folding and unfolding ratio characteristic of the antenna provides bionic improvement inspiration for various urgent properties such as stealth and size reduction of a satellite antenna structure, the shape of the antenna folding and unfolding array can be changed in a multi-stage mode, and the antenna folding and unfolding array is suitable for large-scale popularization and application. Therefore, the functions of optimal frequency band adjustment, optimal area expansion, switching of different scanning ranges and the like are realized.

The invention provides an end-to-end intelligent system development method and system based on an AI large model. The method comprises the following steps: S1, receiving a system demand described by a user through a natural language; s2, analyzing the requirements through an AI large model and converting the requirements into formalized specifications; s3, automatically generating a system architecture and dividing components; s4, generating each component code in a recursive manner; s5, dynamically detecting and repairing the generated code problem; s6, automatically integrating all components to form a complete system; and S7, continuously optimizing the system based on system feedback or user feedback. According to the method, work of multiple stages such as demand analysis, architecture design, module division, code writing and testing is automatically completed, distributed integration and smart contract are performed, distributed integration verification is performed through a block chain smart contract and a graph database, data consistency, interface matching and security among components are ensured, the integration process is optimized, and the integration efficiency is improved. Cross-domain cooperation is achieved, and the development efficiency is high.

The invention provides an assembly line management method, a processor, computer equipment, a medium and a chip, and belongs to the technical field of chips and semiconductors. The method comprises the steps of determining a first assembly line based on a plurality of instructions of the same category; in the process of executing a first instruction based on the first assembly line, for any stage in the multiple stages, when the first instruction is in the stage, detecting whether the first instruction is abnormal or not in the stage; under the condition that the first instruction is abnormal, carrying out retransmission operation on the first instruction; according to the execution process of a second instruction in the first assembly line, the second instruction continues to be executed on the basis of the first assembly line, and the second instruction is an instruction which is executed after the first instruction in the multiple instructions and is not abnormal currently. According to the technical scheme, the execution efficiency of multiple instructions and the utilization rate of an assembly line are improved.

The invention provides a game resource optimization method and device based on a predictive algorithm and a storage medium. The method comprises the following steps: compiling a source code of a game voice tool into a bytecode file and generating an executable game voice tool main body; dividing the loading process of the non-core resources into a plurality of stages, and pre-loading the non-core resources through at least one background thread; generating a target resource list for a subsequent game scene by using a predictive algorithm; executing a loading operation on the target resource which is not loaded by utilizing a background thread, and executing a releasing operation on the loaded resource which is judged to be lower than a preset threshold value in the subsequent game scene by the predictive algorithm, wherein the use probability of the loaded resource is lower than the preset threshold value; a high execution priority is set for a voice communication module, and the loading priority of non-core resources and the task allocation order of background threads are dynamically adjusted according to the real-time condition of voice communication. According to the method and the device, memory resource utilization can be improved, CPU (Central Processing Unit) occupation is reduced, and game jamming caused by resource competition is reduced.

The invention discloses an elastic power distribution network multi-type remote control switch and soft switch combined planning method, which relates to the field of power distribution networks and comprises the following steps: step 1), establishing a target function with the minimum sum of investment cost of a soft switch and a multi-type remote control switch and power loss cost in a fault; 2) establishing a power distribution network operation model considering multiple stages of active island, degradation, fault isolation and fault recovery and protection constraints; and step 3), converting the operation model into a mixed second-order cone integer programming model by adopting a quadratic term linearization method to solve the mixed second-order cone integer programming model. According to the method, multiple actually existing remote control switches and protection constraints are innovatively coordinated in multiple stages, and the capability of the distribution network for resisting disasters is effectively improved.

The invention relates to the technical field of photosensitive dry film production, in particular to a drying device for photosensitive dry film production, which is characterized in that a shell is internally provided with five chambers from left to right in sequence, namely a preheating chamber, a wetting chamber, a drying chamber, a cooling chamber and a winding chamber, a cooling assembly used for cooling the photosensitive dry film is installed in the cooling chamber, and guide rollers used for guiding the photosensitive dry film are installed in the five cavities in the shell, so that the photosensitive dry film gradually moves from the preheating chamber in the preset direction, and winding is completed through a winding device installed in the winding chamber. According to the photosensitive dry film drying device, the preheating chamber, the wetting chamber, the drying chamber and other parts are arranged, the drying effect can be effectively improved through multiple stage type treatment modes, and the situation that in the high-temperature drying process of a photosensitive dry film, due to the fact that water on the surface volatilizes rapidly and water in a film layer volatilizes slowly, the photosensitive dry film warps or deforms is prevented.

Systems, methods, tangible non-transitory computer-readable media, and devices for autonomous vehicle operation are provided. For example, a computing system can receive object data that includes portions of sensor data. The computing system can determine, in a first stage of a multiple stage classification using hardware components, one or more first stage characteristics of the portions of sensor data based on a first machine-learned model. In a second stage of the multiple stage classification, the computing system can determine second stage characteristics of the portions of sensor data based on a second machine-learned model. The computing system can generate an object output based on the first stage characteristics and the second stage characteristics. The object output can include indications associated with detection of objects in the portions of sensor data.

The invention relates to the technical field of PCB machining, and provides a tool compensation machining method and PCB machining device.The tool compensation machining method is used for machining a plate, machining of the plate has multiple stages, and the tool compensation machining method comprises the steps that the initial tool diameter of a tool before machining is detected, and an initial tool compensation value is generated; according to the initial cutter compensation value, a cutter is driven to execute first-stage machining; detecting the actual cutter diameter of the cutter and generating a first cutter compensation value under the condition that the cutter completes the machining of the first stage and the cutter is lifted; and according to the first cutter compensation value, the cutter is driven to execute second-stage machining. According to the method, self-adaptive dynamic compensation of tool abrasion can be achieved, the machining precision is guaranteed, and the machining quality and the automation level are improved.

The invention provides a device design system and a device design method, and relates to the technical field of device design, and the system comprises an obtaining module which is used for obtaining a plurality of pieces of design data related to a plurality of process sections, data processing algorithms corresponding to the plurality of pieces of design data, and an association relationship among the plurality of pieces of design data, the design data is used for indicating hardware parameters and / or process parameters of the device; the processing module is used for determining a plurality of stage design results according to the plurality of design data, the data processing algorithms corresponding to the plurality of design data and the incidence relation among the plurality of design data, the plurality of stage design results correspond to the plurality of process sections, and the plurality of stage design results serve as the basis for preparing the device; the display module is used for displaying a result interface, and the result interface comprises a plurality of stage design results. Based on the scheme, the efficiency and accuracy of device design can be improved.

The invention discloses machining equipment for a low-temperature ethylene pump space guide vane, and relates to the technical field of machining of the low-temperature ethylene pump space guide vane. The device comprises a center supporting electric cylinder, an edge adjusting electric cylinder, a disc plate and a displacement assembly, and automatic and accurate adjustment of the spatial orientation and angle of the guide vane is achieved; guide vanes of different specifications can be efficiently clamped through the double-three-jaw chuck structure; the cutting device is provided with inner and outer reinforcing sleeves to improve the rigidity of the cutter and prevent deformation; double motors are adopted to drive planetary gears for transmission, stepless speed change of the cutter is achieved, and the machining requirements of different materials are met; the multiple stages of displacement assemblies work cooperatively, the axial position, the radial position and the rotating position of the guide vane are flexibly controlled, and efficient and high-precision machining of a complex curved surface is completed through multi-degree-of-freedom linkage.

The invention discloses a waste battery raw material refining device which comprises a box body, a filter plate is arranged in the box body, a multi-stage crushing mechanism is arranged above the filter plate, and the multi-stage crushing mechanism comprises a rotary cutting assembly and an extrusion crushing assembly; the rotary cutting assembly comprises a cutting motor, the cutting motor is installed on the outer side wall of the box body, the output end of the cutting motor is connected with a rotating shaft, and a plurality of cutting blade sets are arranged on the rotating shaft. The extrusion crushing assembly comprises a driving extrusion plate and a driven extrusion plate which are oppositely arranged, the driving extrusion plate is connected to the side wall of the box body through a telescopic air cylinder, the driven extrusion plate is connected with the side wall of the box body through a telescopic rod, and in addition, the front end face of the driving extrusion plate and the front end face of the driven extrusion plate are provided with staggered crushing teeth; and ultrasonic vibration assemblies are arranged in the two. Through the synergistic effect of the multiple stages of crushing mechanisms, crushed battery materials are smaller in granularity and more uniform, the dissolving speed is increased, and the raw material refining efficiency is improved.