Numerical control machine tool for composite machining of steel structure

Abstract

The application discloses a numerical control machine tool for steel structure composite machining, which comprises a chassis, a Y-axis sliding saddle, an X / Y / Z-axis driving assembly, a bearing table, a clamping assembly, a portal frame, a machining sliding frame and three execution assemblies, and a controller comprises a model import and feature recognition module, a machining sequence planning module, a path generation and simulation module, a real-time monitoring module, a self-adaptive compensation module and a cooperative control module; the controller can automatically analyze a three-dimensional model and extract machining features, optimize machining sequences and parameters through a genetic algorithm, and generate a collision-free tool path; vibration, cutting force and temperature signals are collected in real time during the machining process, and a fuzzy neural network is used to dynamically adjust a feed speed and a spindle speed; switching is realized through a state machine between each process, and online detection and local rework functions are provided; the application realizes full-automatic intelligent machining from a three-dimensional model to a finished product, and improves the efficiency, precision and safety of steel structure composite machining.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool technology, specifically to a CNC machine tool for composite machining of steel structures. Background Technology

[0002] Steel structural components are widely used in construction, bridges, and machinery manufacturing. Their processing typically involves multiple steps, including blanking, drilling, milling, and grinding. In traditional processing methods, different steps often need to be completed on different machine tools. Repeated clamping of the workpiece leads to decreased positioning accuracy, low production efficiency, and increased labor costs and equipment footprint.

[0003] To address these issues, several multi-functional machining centers, such as drilling-milling centers and turning-milling centers, have emerged in the market. However, most existing multi-functional machining center control systems employ simple sequential control methods, lacking collaborative optimization between machining units. Machining parameters rely on operator experience for setting and are difficult to adjust in real-time based on the actual workpiece condition and tool wear, resulting in unstable machining quality and short tool life. Furthermore, for complex contours and irregular hole systems commonly found in steel structural components, existing control systems lack intelligent path planning and collision detection functions, making programming complex and prone to interference accidents.

[0004] Therefore, it is necessary to invent a CNC machine tool for composite steel structure processing to solve the above problems. Summary of the Invention

[0005] To address the technical problems in the prior art, this application provides a CNC machine tool for composite machining of steel structures, including a base frame, a Y-axis slide saddle, X / Y / Z-axis drive components, a support table, a clamping component, a gantry, a machining carriage, and three execution components. The control unit includes a model import and feature recognition module, a machining sequence planning module, a path generation and simulation module, a real-time monitoring module, an adaptive compensation module, and a collaborative control module. The control unit can automatically parse the 3D model and extract machining features, optimize the machining sequence and parameters through a genetic algorithm, and generate a collision-free tool path. During machining, vibration, cutting force, and temperature signals are collected in real time, and the feed rate and spindle speed are dynamically adjusted using a fuzzy neural network. Switching between processes is achieved through a state machine, and online detection and local rework functions are provided.

[0006] The CNC machine tool for composite steel structure processing provided in this application adopts the following technical solution: it includes a base frame, and a Y-axis slide saddle is provided on the top of the base frame. The Y-axis slide saddle is equipped with a Y-axis moving component for driving the Y-axis slide saddle to adjust along the Y-axis direction. A support platform is slidably disposed above the Y-axis slide saddle, and the support platform is equipped with an X-axis moving component for driving the support platform to adjust along the X-axis direction; Clamping assemblies are installed on both sides above the support platform for clamping the workpiece to be processed. The base frame is equipped with gantry frames on both sides, and a processing slide is provided on one side of the gantry frame. The processing slide is equipped with a Z-axis lifting assembly for driving the processing slide to adjust along the Z-axis direction, and the Z-axis lifting assembly is mounted on the gantry frame. The top of one side of the processing carriage is equipped with an execution component one, the right side is equipped with an execution component two, and the left side is equipped with an execution component three. A control component is installed on one side of the gantry frame, and the control component includes: The model import and feature recognition module is used to receive the 3D model file of the workpiece to be processed and automatically extract hole features, planar features and curved surface features; The processing sequence planning module, connected to the feature recognition module, is used to generate a processing sequence of drilling, milling, and grinding based on the extracted features, and to plan the calling order and processing parameters of each execution component. The path generation and simulation module is used to generate the tool motion trajectory for each machining process, perform interference and collision detection, and output a collision-free machining path. The real-time monitoring module is used to collect signals from vibration sensors, cutting force sensors, and temperature sensors during the machining process; The adaptive compensation module dynamically adjusts the feed rate, spindle speed, and depth of cut based on data collected by the real-time monitoring module using a fuzzy neural network model. The collaborative control module is used to synchronously control the X-axis movement component, Y-axis movement component, Z-axis lifting component, and execution component one, execution component two, and execution component three.

[0007] The present invention has the following beneficial effects: 1. This invention, by setting up a model import and feature recognition module, can automatically parse 3D models and extract processing features without the need for manual programming, which greatly reduces the technical requirements for operators and avoids the defects of easy errors in manual programming. 2. This invention integrates a real-time monitoring module and an adaptive compensation module, and uses a fuzzy neural network to dynamically adjust the processing parameters, which can effectively cope with time-varying factors such as uneven workpiece material and tool wear, and ensure the stability of processing quality. 3. The present invention has a built-in path generation and simulation module and a collision detection algorithm, which eliminates the risk of interference before processing. It is particularly suitable for processing complex contours of steel structures and avoids tool collision accidents caused by programming negligence. 4. This invention achieves online detection and partial rework between processes through a processing quality self-inspection module, forming a closed-loop control of "processing-detection-compensation", which improves the finished product qualification rate. Attached Figure Description

[0008] Figure 1 This is a three-dimensional structural diagram of the present invention; Figure 2 This is a three-dimensional structural diagram of the present invention from another perspective; Figure 3 This is a diagram showing the overall architecture of the control module of the present invention; Figure 4 This is a schematic diagram of the three-layer collision detection and obstacle avoidance logic of the present invention; Figure 5 This is a decision logic diagram of the adaptive compensation module of the present invention; Explanation of reference numerals in the attached drawings: 1. Base frame; 2. Y-axis slide saddle; 21. Y-axis moving assembly; 22. X-axis moving assembly; 23. Support platform; 24. Clamping assembly; 3. Gantry frame; 4. Machining carriage; 41. Z-axis lifting assembly; 42. Actuation component one; 43. Actuation component two; 44. Actuation component three; 5. Control component. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The CNC machine tool for composite processing of steel structures involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] Please see Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5 The CNC machine tool for composite steel structure processing shown includes a base frame 1, with a Y-axis slide saddle 2 disposed above the base frame 1. The Y-axis slide saddle 2 is equipped with a Y-axis moving component 21. In this embodiment, the Y-axis moving component 21 is used to drive the Y-axis slide saddle 2 to move horizontally along the Y-axis direction. It should be noted that the Y-axis moving component 21 can adopt a combination structure of a lead screw and nut pair, linear guide rail and servo motor, which are conventional in the art. Its specific model and installation method are common knowledge in the art and will not be described in detail here. A support platform 23 is slidably disposed above the Y-axis slide saddle 2, and the support platform 23 is equipped with an X-axis moving component 22; the X-axis moving component 22 is used to drive the support platform 23 to adjust horizontally along the X-axis direction; similarly, the specific implementation of the X-axis moving component 22 (such as a linear motor, gear rack, etc.) is a technical means that can be conventionally selected by those skilled in the art according to the processing accuracy and load requirements, and this application does not make any special limitations; Clamping components 24 are installed on both sides above the bearing platform 23. The clamping components 24 are used to clamp and fix the workpiece to be processed. The clamping components 24 can be pneumatic clamps, hydraulic clamps or manual quick clamps. Their structure and working principle are mature technologies in the field. Only functional description is given here. The base frame 1 is equipped with gantry frames 3 on both sides, which span across the Y-axis slide saddle 2 and the support platform 23. A machining slide 4 is provided on one side of the gantry frame 3, and the machining slide 4 is equipped with a Z-axis lifting assembly 41. The Z-axis lifting assembly 41 is installed on the gantry frame 3 and is used to drive the machining slide 4 to lift and adjust along the Z-axis direction (vertical direction). The Z-axis lifting assembly 41 can adopt conventional ball screw lifting mechanism, linear guide rail and counterweight balancing system in the art. The specific selection and assembly of these components are conventional designs in the art. The machining carriage 4 has an execution component 1 42 mounted on one side top, an execution component 2 43 mounted on the right side, and an execution component 3 44 mounted on the left side. It should be noted that the execution components 1 42, 2 43, and 3 44 are used to perform different machining processes, specifically drilling, milling, and grinding. Their specific forms (such as electric spindles, power heads, tool magazine interfaces, etc.) are all machining units known in the art. This application only limits their installation position relationship and does not involve improvements to the specific internal structure. A control component 5 is installed on one side of the gantry frame 3; the control component 5 includes: The model import and feature recognition module is used to receive the 3D model file of the workpiece to be processed and automatically extract hole features, planar features and curved surface features; Furthermore, in the above technical solution, the model import and feature recognition module adopts a dual-channel parallel feature recognition architecture: The first channel is a processing feature classification channel based on graph neural network. Each face of the 3D model is treated as a node in the graph, and the adjacency relationship between faces is treated as an edge. An attribute adjacency graph is constructed. The attribute adjacency graph is processed by a spectral domain-spatial domain hybrid graph convolutional network to identify through holes, threaded bottom holes, countersunk holes, waist-shaped holes, right-angled planes and arc surfaces, and output the spatial pose and geometric parameters of each feature. The second channel is a fine feature segmentation channel based on point cloud deep learning. It uniformly samples the surface of the 3D model into point cloud data, performs semantic segmentation on the point cloud through the PointNet++ network, and classifies each point into regions, including drilling regions, milling regions, grinding regions and non-processing regions. The recognition results from the two channels are merged by a feature fusion layer to generate a structured feature database containing feature ID, feature type, spatial pose, geometric parameters, tolerance level, and surface roughness requirements.

[0011] Furthermore, in the above technical solution, the processing sequence planning module includes a multi-objective optimizer based on the NSGA-II non-dominated sorting genetic algorithm. The decision variables of the multi-objective optimizer are the process arrangement vector and the continuous processing parameters of each process. The optimization objective is to minimize the total processing time T. total and minimize tool wear W total ; The total processing time T total =Σ(t move，i +t machine，i +t toolchange，i ), where t move，i The idle travel time between adjacent workstations, t machine，i For actual cutting time, t toolchange，i This is the tool change time; The tool wear W total =Σ(k material ×k tool ×L i ), where k material For the wear coefficient of the workpiece material, k tool For the wear coefficient of the tool material, L i Let be the cutting path length of the i-th process; The constraints of the multi-objective optimizer include: available rotational speed range constraints for each execution component, maximum allowable feed rate constraints, maximum allowable cutting force constraints, spindle power constraints, process sequence constraints between operations, and geometric reachability constraints between the execution component and the workpiece; The multi-objective optimizer outputs a Pareto front solution set after population iterative evolution, and the inflection point solution is selected by the operator or automatically selected by the system as the final optimization processing sequence.

[0012] The processing sequence planning module, connected to the feature recognition module, is used to generate a processing sequence of drilling, milling, and grinding based on the extracted features, and to plan the calling order and processing parameters of each execution component. The path generation and simulation module is used to generate the tool motion trajectory for each machining process, perform interference and collision detection, and output a collision-free machining path. Furthermore, in the above technical solution, the path generation and simulation module incorporates a gantry machine tool forward kinematics model and an S-shaped acceleration / deceleration planner, converting the tool path into motion commands for each servo axis, and employing a three-layer progressive collision detection strategy: The first layer is a coarse axial bounding box (AABB) test, which encloses each object in an axial bounding box and tests the intersection. The second layer involves detection within the directional bounding box (OBB). For objects that intersect within the AABB, the directional bounding box is used to perform more accurate intersection detection. The third layer is based on the Separated Axis Theorem (SAT) for fine detection of triangular facets. For object pairs where OBB still intersects, the triangular mesh model is extracted and interference is detected on a triangular facet-by-triangular facet basis. If interference is detected, the path generation and simulation module automatically triggers obstacle avoidance path replanning: randomly sample collision-free poses in the configuration space near the interference point, and use RRT* to quickly explore the random tree star topology algorithm to connect the blocked trajectories. If the number of retries exceeds the preset threshold and a collision-free path still cannot be found, the process is marked as requiring manual intervention and a prompt is issued.

[0013] The real-time monitoring module is used to collect signals from vibration sensors, cutting force sensors, and temperature sensors during the machining process; Furthermore, in the above technical solution, the real-time monitoring module includes a triaxial accelerometer, a strain-type triaxial cutting force sensor, a thermocouple temperature sensor, a spindle current Hall effect sensor, an eddy current displacement sensor, an FPGA coprocessor, and an ARM processor. The signals from the vibration sensor and the cutting force sensor are digitized by a 24-bit Δ-Σ ADC and then preprocessed by an FIR digital filter by the FPGA coprocessor. The FIR digital filter includes a fourth-order Butterworth bandpass filter (passband 0.5~500Hz) for the vibration signal and a fourth-order Butterworth low-pass filter (cutoff frequency 200Hz) for the cutting force signal. The preprocessed data is written to the shared memory ring buffer on the ARM processor side via DMA direct memory access. Each data packet is accompanied by a hardware timestamp for the adaptive compensation module to read.

[0014] The adaptive compensation module dynamically adjusts the feed rate, spindle speed, and depth of cut based on data collected by the real-time monitoring module using a fuzzy neural network model. Furthermore, in the above technical solution, the fuzzy neural network model in the adaptive compensation module is a five-layer adaptive neurofuzzy inference system (ANFIS) architecture: The first layer is the input layer, which receives four input variables: the normalized effective value of vibration amplitude, the average value of cutting force, the temperature deviation, and the current feed rate. The second layer is the membership function layer, where each input variable corresponds to three Gaussian membership functions, representing the low, medium and high levels of state respectively. The third layer is the fuzzy rule layer, which calculates the trigger strength of each rule by multiplying the membership degrees of each input. The consequent of the rule is a first-order linear function. The fourth layer is the normalization layer, which calculates the normalized trigger strength of each rule; The fifth layer is the output layer, which outputs the feed rate correction coefficient k. v(Range 0.6~1.2) and spindle speed correction factor k s (Range 0.7~1.1); The antecedent and consequent parameters of the fuzzy neural network model are continuously updated through a combination of offline training and online incremental learning. During the offline training phase, a hybrid learning algorithm is used for initial training with historical processed data; During the online processing phase, after each workpiece is processed, the currently collected sensor data and processing quality data are used as new samples to fine-tune the network parameters through backpropagation in a small batch manner.

[0015] Furthermore, in the above technical solution, the adaptive compensation module adjusts the parameters according to the following decision logic: When k v ≥0.9 and k s When the value is ≥0.95, maintain the current processing parameters; When 0.8≤k v <0.9 or 0.85≤k s When the value is less than 0.95, reduce the feed rate or spindle speed by 0.95 times. When 0.7≤k v <0.8 or 0.75≤k s When the value is less than 0.85, reduce the feed rate or spindle speed by 0.9 times. When k v <0.7 or k s If the value is less than 0.75 and persists for more than three control cycles, a tool change warning will be triggered and machining will be paused. When k v When the feed rate is greater than 1.1 and the current cutting force is less than 60% of the rated value, the feed rate is increased by 1.05 times until the upper limit is reached.

[0016] The collaborative control module is used to synchronously control the X-axis moving component 22, the Y-axis moving component 21, the Z-axis lifting component 41, and the execution component one 42, the execution component two 43, and the execution component three 44.

[0017] Furthermore, in the above technical solution, the collaborative control module incorporates a hierarchical finite state machine, and the top-level master state machine includes the following states: S0 Standby, S1 Clamping Detection, S2 Zeroing, S3 Drilling Process, S4 First Tool Change Position Adjustment, S5 Milling Process, S6 Second Tool Change Position Adjustment, S7 Grinding Process, S8 Machining Completed, S9 Pause and S10 Emergency Stop. The state machine automatically jumps according to the process list output by the processing sequence planning module, and the state transition is driven by external events, internal events and communication events; The main states of the drilling, milling and grinding processes are each equipped with a sub-process state machine. The sub-process state machine includes positioning, rapid advance, feed, pause, tool retraction and position change sub-states. Each sub-state corresponds to a set of axis motion commands and I / O operations. All state transitions are recorded in the system log with timestamps.

[0018] Furthermore, in the above technical solution, the control component 5 also includes a processing quality self-inspection module, which integrates a contact radio probe on the execution component 1 42 and a line laser profile sensor on the execution component 2 43. The contact-type wireless probe is used to measure the diameter and depth of no less than three drilled holes after the drilling process is completed. If the measurement deviation exceeds the first preset threshold, the local rework process is automatically triggered: the control execution component 42 returns to the out-of-tolerance hole position, calculates the coordinate compensation amount and depth compensation amount according to the deviation value and performs secondary drilling. The maximum number of compensations for the same hole position is three. An alarm is issued after the maximum number of compensations is exceeded. The line laser contour sensor is used to scan the machining plane after the milling process to generate point cloud data, compare it with the theoretical plane to calculate the flatness error, and if the flatness error exceeds the second preset threshold, a local compensation path is generated according to the error distribution map to drive the execution component 43 to mill the out-of-tolerance area again. The processing quality self-inspection module also includes a statistical process control unit, which automatically records the measurement data of each workpiece and generates a mean-range control chart. When seven consecutive data points are on the same side of the mean or three consecutive data points exceed the 2σ limit, a process deviation warning is issued.

[0019] The following is a detailed description of the workflow of each module included in control component 5 in this embodiment during the specific processing.

[0020] The model import and feature recognition module operates as follows: Operators import the STEP (AP203 / AP214) format 3D model file of the steel structure to be processed into control component 5 via USB flash drive or Ethernet. After importing the model, the system first uses the Open Cascade geometric kernel to analyze the B-rep boundary representation data of the model, extracting the topological information of the face, edge, and vertex. For complex models with more than 500,000 triangular faces, the system automatically executes the Quadrature Mesh Reduction (QEM) algorithm to compress the number of faces to less than 200,000, while ensuring that the geometric accuracy error of the feature region is less than 0.01 mm.

[0021] Subsequently, the feature recognition module initiates dual-channel parallel recognition: The first channel treats each face of the 3D model as a node in the graph, and the adjacency relationship between faces as edges in the graph, constructing an attribute adjacency graph. The attribute adjacency graph is processed by a spectral-spatial hybrid graph convolutional network pre-trained with 100,000 sets of labeled data to identify through holes, threaded bottom holes, countersunk holes, waist-shaped holes, right-angled planes, and arc surfaces, and outputs the spatial pose and geometric parameters of each feature; The second channel uniformly samples the model surface into 10,000 point cloud data points, and performs semantic segmentation on the point cloud using the PointNet++ network, classifying each point into drilling areas, milling areas, grinding areas, or non-machined areas. The recognition results from the two channels are merged through a feature fusion layer. When the two channels classify the same feature inconsistently, the result of the first channel takes precedence, but the confidence level of the second channel is used as a weighting reference for subsequent processing parameter settings. This ultimately generates a structured feature database. Each record in the database includes: feature ID (globally unique code, such as "F_0001"), feature type (through hole / threaded bottom hole / countersunk hole / waist hole / plane / curved surface), spatial pose (position coordinates x, y, z and orientation Euler angles α, β, γ), geometric parameters (diameter, depth, length, width, etc.), tolerance level (extracted from the model PMI annotation; if no annotation is provided, IT10 defaults are used), and surface roughness requirements (Ra value). The feature recognition module completes all feature recognition from model import to display on the user interface.

[0022] After the operator confirms the feature recognition results are correct on the user interface, they click the "Automatic Planning" button, and the processing sequence planning module is immediately started. This module embeds a multi-objective optimizer based on the NSGA-II non-dominated sorting genetic algorithm. The optimizer first reads all feature records from the structured feature database and establishes a complete processing constraint model, which includes the following four aspects: (i) Tool Library Constraints: The system has a built-in tool database that records the specifications, available speed range, maximum feed rate, maximum depth of cut, and remaining tool life of all currently available tools. For example, the available speed range of a Φ10mm high-speed steel drill bit is 800~5000rpm, and the maximum feed rate is 200mm / min; the available speed range of a Φ16mm carbide flat end mill is 1000~10000rpm, the maximum feed rate is 500mm / min, and the maximum depth of cut is 3mm; the available speed range of a Φ25mm ceramic fiber grinding head is 500~3000rpm, and the maximum feed rate is 300mm / min.

[0023] (ii) Process sequence constraints: including: before drilling, a flat guide recess must be milled on the workpiece surface by the execution component 2 43 to prevent the drill bit from slipping; the threaded bottom hole must be chamfered immediately after drilling; the grinding process must be carried out after all milling processes are completed to prevent the grinding surface from being scratched by subsequent processing; the plane milling should be carried out in the order of rough milling → semi-finish milling → finish milling.

[0024] (iii) Geometric reachability constraint: For each feature, the system calculates whether execution component 1 42, execution component 2 43, and execution component 3 44 will spatially interfere with the workpiece, fixture, or gantry 3 when approaching the feature from different directions. If a feature can only be processed by a specific execution component, then the constraint is an inviolable hard constraint.

[0025] (iv) Cutting physical constraints: including the maximum allowable cutting force of each actuator (default 2000N, determined by the force sensor range) and the maximum spindle power (default 15kW).

[0026] Under the above constraints, the optimizer sets the decision variables as the process arrangement vector π = (π1, π2, ..., π). N (where N is the total number of features, π) i The value is {drilling, milling, grinding} × {feature ID}, and the feed rate v for each process. f Spindle speed n and depth of cut a p Three continuous parameters. The optimization objective is to simultaneously minimize the total processing time T. total =Σ(t move，i +t machine，i +t toolchange，i and minimize tool wear W total =Σ(k material ×k tool ×L i ), where t move，i The idle travel time between adjacent workstations, i.e., the time required for the actuator to move from the end position of the previous workstation to the start position of the current workstation in the i-th operation, is calculated from the maximum traversal speed and traversal distance of the X-axis, Y-axis, and Z-axis based on the machine tool kinematics model. machine，i The actual cutting time, i.e., the actual cutting time of the i-th operation, is equal to the cutting path length divided by the feed rate and t. toolchange，i The tool change time is the time taken before the start of the i-th operation. If switching the execution component (i.e., changing the machining tool unit) is required, a tool change time is generated, which is a fixed constant of 5 seconds. If switching the execution component is not required, then t... toolchange，i =0,k material For the wear coefficient of the workpiece material, take 1.0 for Q235 steel, 1.3 for 45 steel, and 1.8 for stainless steel; k toolFor tool material wear coefficients, 1.2 is used for high-speed steel and 0.8 for cemented carbide; L i Let be the cutting path length of the i-th process.

[0027] The optimizer runs the NSGA-II algorithm with a population size of 200, a maximum number of iterations of 500 (the population size and number of iterations are chosen based on the following criteria: when the population size is below 100, the algorithm is prone to getting trapped in local optima; when it is above 300, the computation time increases significantly; when the number of iterations is below 300, the Pareto front has not yet converged; when it is above 800, computational resources are wasted), a crossover probability of 0.8, and a mutation probability of 0.05. After 500 generations of evolution, the optimizer outputs a Pareto front solution set (usually containing 20 to 50 non-dominated solutions). The user interface displays the estimated machining time and tool wear of each solution in the form of a scatter plot, which is then selected and confirmed by the operator. If the operator does not make a selection within 60 seconds, the system automatically selects the solution with the largest inflection point curvature on the Pareto front as the final optimized machining sequence.

[0028] After the optimized machining sequence is confirmed, the path generation and simulation module automatically starts. This module generates the corresponding toolpath based on the type of each process: for drilling, a pecking drill cycle (G83 standard cycle) is used, with each pecking drill depth being 0.3 times the tool diameter and a retraction height of 0.5 mm. For deep holes with a depth greater than 3 times the diameter, an additional chip removal action is performed every 10 mm of drilling depth to the hole opening; for milling, planar milling uses the equal residual height method to generate the cutting path, with the line spacing calculated using the formula: "Line spacing = 2 × (tool radius)". 2 -(Tool radius - Allowable residual height) 2 ) 1 / 2 "Calculations show that freeform surface milling uses the isoparametric method to generate tool paths, with tool contact points evenly distributed along the U / V direction of the surface; for the grinding process, a reciprocating trajectory is generated by offsetting 0.5mm along the normal of the workpiece contour, with an overlap rate of 30% between adjacent trajectories, and the axis of the grinding head is always kept consistent with the normal direction of the workpiece surface."

[0029] After the trajectory is generated, the module calls the built-in forward kinematics model of the gantry machine tool to convert the tool path (in the workpiece coordinate system) into motion commands (in the machine coordinate system) for three servo axes: X-axis traverse component 22, Y-axis traverse component 21, and Z-axis lifting component 41. The kinematic model considers the travel of each axis (X-axis 0~2000mm, Y-axis 0~1500mm, Z-axis 0~800mm) and the maximum acceleration (5m / s²). 2 ) and maximum jerk (50m / s²) 3 It also employs an S-type acceleration / deceleration planner to ensure continuous acceleration variation, reducing mechanical shock and vibration.

[0030] Subsequently, the module employs a three-layer progressive strategy to perform collision detection: The first layer is coarse detection using axial bounding boxes (AABB)—the workpiece, fixture, gantry 3, machining carriage 4, and each execution component are all enclosed in axial bounding boxes, and each bounding box is checked for intersection in each interpolation cycle. If they do not intersect, the subsequent detection is skipped. If they intersect, the second layer, directional bounding boxes (OBB), is used for detection—object pairs that intersect using AABB are further checked for intersection using directional bounding boxes. OBB takes into account the actual orientation of the objects, which can reduce the false alarm rate by about 70%. If the OBBs still intersect, the third layer, fine detection based on the separating axis theorem (SAT), is used—the triangular mesh model of the intersecting object pairs is extracted, and interference is detected on a triangular facet basis. If no collision is detected after three layers of detection, the module outputs the collision-free processing code and stores it in the cache. If interference is detected, the module automatically triggers obstacle avoidance path replanning—randomly sampling collision-free poses in the configuration space (C-space) near the interference point, and using RRT* (Fast Explore Random Tree Star Algorithm) to connect the blocked trajectory. The algorithm has a maximum of 1000 retries. If a collision-free path cannot be found after 1000 retries, the process is marked as "requiring manual intervention" on the user interface, and an audio-visual prompt is issued, along with the specific interference location and cause. The entire path generation and simulation process is completed within 1 to 3 minutes.

[0031] After the collision-free machining code is generated, the operator places the steel structure workpiece to be machined on the support platform 23 and presses the "clamp" button, at which point the collaborative control module is activated. The collaborative control module has a built-in hierarchical finite state machine, whose top-level master state machine includes the following states: S0 standby, S1 clamping detection, S2 zeroing, S3 drilling operation, S4 first tool change position adjustment, S5 milling operation, S6 second tool change position adjustment, S7 grinding operation, S8 machining completed, S9 pause, and S10 emergency stop.

[0032] The specific execution flow of the collaborative control module is as follows: First, the state machine jumps from the S0 standby state to the S1 clamping detection state—the controller 5 sends an energizing command to the solenoid valve of the clamping assembly 24, and the pneumatic clamp begins to clamp the workpiece. At the same time, the pressure sensor installed on the clamping assembly 24 provides real-time feedback of the clamping force value at a sampling rate of 100Hz. When the clamping force reaches the preset threshold of 500N and remains stable for 1 second, the state machine determines that the clamping detection is passed and jumps to the S2 zeroing state.

[0033] In the S2 zeroing state, the collaborative control module simultaneously sends zeroing commands to the servo drives of the Y-axis moving component 21, X-axis moving component 22, and Z-axis lifting component 41 via the EtherCAT bus. The three motion axes move towards their respective zero-point switches at rapid traversal speeds (20000 mm / min for X / Y axes and 10000 mm / min for Z-axis). After touching the zero-point switch, they decelerate to a crawling speed (500 mm / min for each axis) and continue moving until the zero-point signal is triggered. Then, each axis moves in the opposite direction to the zero-point offset position (5 mm offset for each axis) to complete precise zeroing. When the zero-point signals of all axes are triggered and the position feedback error is less than 0.001 mm, the state machine determines that zeroing is complete and jumps to the S3 drilling process.

[0034] In the S3 drilling process state, the collaborative control module reads all machining instructions for the drilling process from the machining code cache and executes them cyclically according to the following sub-state machine: First, it enters the "positioning" sub-state—the X-axis moving component 22 and the Y-axis moving component 21 move the workpiece on the support table 23 at a rapid traverse speed to a position where the current drilling position is within ±0.01mm directly below the execution component 42; after positioning, it enters the "rapid traverse" sub-state—the Z-axis lifting component 41 drives the machining carriage 4 to descend at a speed of 2000mm / min, so that the drill tip of the execution component 42 reaches a safe height of 2mm from the workpiece surface; then it enters the "feed" sub-state. —The spindle of the execution component 42 begins to rotate at an optimized speed (e.g., 2500 rpm for a Φ10mm drill bit on Q235 steel). The Z-axis lifting component 41 drives the drill bit to cut into the workpiece at an optimized feed rate (e.g., 150mm / min). Simultaneously, the real-time monitoring module collects vibration and cutting force signals at a 2kHz sampling rate. When the drill bit reaches the preset depth, the Z-axis lifting component 41 quickly retracts to a safe height at a speed of 1000mm / min. Then, it enters the "repositioning" sub-state—determining whether there are any undrilled holes. If so, it returns to the "positioning" sub-state to continue drilling the next hole. If all holes have been drilled, it exits the S3 drilling process sub-state machine. After each hole is drilled, the sub-state machine of this process briefly enters the "pause" sub-state (lasting 50ms) to allow the real-time monitoring module to record the peak cutting force and vibration amplitude data of that hole.

[0035] After all drilling operations are completed, the state machine jumps from S3 to S4, the first tool change position adjustment state. The collaborative control module first raises the machining carriage 4 to the highest safe position of the Z-axis travel (more than 300mm from the workpiece surface) via the Z-axis lifting assembly 41. Then, it moves the support table 23 to the tool change station (located at a specific coordinate position on one side of the gantry 3) via the X-axis moving assembly 22 and the Y-axis moving assembly 21. Subsequently, it rotates the execution assembly 43 to the working position via the rotation mechanism on the machining carriage 4. Once the position feedback signal of the execution assembly 43 confirms that it is in place and the Z-axis has reached the safe height, the state machine jumps to S5, the milling operation.

[0036] In the S5 milling process state, the collaborative control module executes all milling operations according to a sub-state machine similar to that of the drilling process—including a cycle of positioning, rapid traverse, feed, pause, and repositioning. The differences are: (i) the milling process calls execution component 243; (ii) the feed rate for milling is slower (e.g., 300 mm / min for planar milling); and (iii) the milling path is a continuous motion trajectory rather than fixed-point machining. Therefore, the "positioning" sub-state loads all trajectory points sequentially into the motion control buffer, and continuous interpolation motion is executed in conjunction with the X / Y / Z axes. During milling, the real-time monitoring module continuously collects sensor signals and sends them to the adaptive compensation module.

[0037] After all milling operations are completed, the machining quality self-inspection module performs online measurement. Once the measurement is satisfactory, the state machine jumps from S5 to S6, the second tool change position adjustment state—the collaborative control module raises the machining carriage 4 to a safe height, moves the support platform 23 to the tool change position, and rotates the machining carriage 4 to rotate the execution component 44 to the working position. After reaching the desired position, the state machine jumps to S7, the grinding operation.

[0038] In the S7 grinding process, the collaborative control module invokes execution component 344 to run the grinding trajectory along the workpiece contour at a relatively low feed rate (e.g., 100 mm / min) and a low rotation speed (e.g., 1500 rpm). During grinding, the real-time monitoring module focuses on monitoring the feedback from the thermocouple temperature sensor. When the temperature of the contact area between the grinding head and the workpiece exceeds 60°C, the collaborative control module automatically activates the coolant solenoid valve to spray coolant, while simultaneously reducing the feed rate by 30%.

[0039] After all grinding processes are completed, the state machine jumps to the S8 machining completion state—the collaborative control module raises the machining carriage 4 to its highest safe position via the Z-axis lifting assembly 41, then moves the support platform 23 to the loading / unloading station (i.e., the zero point position of the X and Y axes) via the X-axis moving assembly 22 and the Y-axis moving assembly 21, and finally sends a power-off command to the solenoid valve of the clamping assembly 24, causing the pneumatic clamp to release the workpiece. At this point, the entire machining process is complete.

[0040] Throughout all the aforementioned processing steps, the real-time monitoring module remains operational. The module is equipped with the following sensors: a triaxial piezoelectric accelerometer mounted on the spindle bearing housings of actuator component 1 (42) and actuator component 2 (43); a strain-type triaxial cutting force sensor mounted at the connection between the Z-axis slide and the machining carriage 4; a K-type thermocouple temperature sensor mounted on the grinding head bracket of actuator component 3 (44); Hall effect spindle current sensors mounted at the outputs of each spindle servo drive; and an eddy current displacement sensor mounted on the Z-axis leadscrew nut housing. All analog sensor signals are digitized by a 24-bit Δ-Σ ADC and then sent to the FPGA coprocessor. The FPGA coprocessor performs preprocessing on the vibration signal using a fourth-order Butterworth bandpass filter (passband 0.5~500Hz, stopband attenuation ≥60dB) and on the cutting force signal using a fourth-order Butterworth low-pass filter (cutoff frequency 200Hz). The preprocessed data is written to the shared memory circular buffer (depth 2048) on the ARM processor side via DMA direct memory access. Each data packet is accompanied by a hardware timestamp with a precision of 1μs, which is read by the adaptive compensation module at 50ms intervals.

[0041] The adaptive compensation module incorporates a five-layer adaptive neural fuzzy inference system (ANFIS). The first input layer receives four input variables: the normalized effective value of the vibration amplitude x1, the average cutting force x2, the temperature deviation x3 (the difference relative to the optimal operating temperature of 20℃), and the current feed rate x4. In the second membership function layer, each input variable corresponds to three Gaussian membership functions (labeled low, medium, and high), with the center value c and width σ of the Gaussian function being trainable parameters. The third fuzzy rule layer contains 81 nodes (3... 4 =81), each node corresponds to a fuzzy rule, the rule form is "IF x1 is A1^j AND x2 is A2^k AND x3 is A3^l AND x4 is A4^m THEN y=f(x1, x2, x3, x4)", where j, k, l, and m take values ​​of 1, 2, and 3 respectively (corresponding to low, medium, and high levels), for a total of 81 combinations. The trigger strength of each rule is calculated by the product of the four input membership degrees, and the consequent function f is a first-order linear function f i =p i ·x1+q i ·x2+r i ·x3+s i ·x4+t i The fourth normalization layer calculates the normalized trigger strength of each rule. ;in Let represent the normalized trigger strength of the i-th rule, with a value in the range (0, 1], and the sum of the normalized trigger strengths of all rules satisfies . w i The original trigger strength of the i-th rule is represented by the product of the four membership functions; w j This represents the original trigger strength of the j-th rule, where j is the summation index variable, iterating from 1 to 81; the fifth output layer outputs the feed rate correction coefficient k. v (Range 0.6~1.2) and spindle speed correction factor k s (Range 0.7~1.1), the correction factor limit value is set based on the safety margin requirements of the machine tool spindle and servo drive, and the output value is the weighted sum of the outputs of each rule successor. .

[0042] It should be further explained that, since the output layer of this neural fuzzy system contains two output variables: the feed rate correction coefficient k... v and spindle speed correction factor k s p, q, r, s, and t actually correspond to two independent linear functions. That is, for k... v and k s Each component has its own independent set of weighting coefficients and bias terms. In this embodiment, the initial values ​​of each weighting coefficient are set according to physical laws: vibration amplitude (x1) and cutting force (x2) have the greatest impact on machining quality, so the absolute values ​​of their corresponding weights p and q are initially set relatively high (e.g., greater than 0.3); temperature deviation (x3) only has a significant impact in the grinding process, so its weight r can be adaptively adjusted in different processes; feed rate (x4) serves as a feedback term for the current state, and its weight s is set to a moderate value (approximately 0.2). The bias term t is initially set to 1.0, representing the baseline correction value when there is no external interference. The above initial values ​​are further optimized using historical data during subsequent offline training, rather than relying entirely on random initialization, to ensure the model's convergence speed and generalization ability.

[0043] Calculation example: Assume the current real-time physical quantity, after normalization, is as follows: x1 (vibration) = 0.6, x2 (cutting force) = 0.8, x3 (temperature deviation) = 0.5, x4 (feed rate) = 0.7.

[0044] Now let's look at rule 5. The definition of this rule is: "IF x1 is high AND x2 is high AND x3 is middle AND x4 is high".

[0045] Step 1: Check the Gaussian function output (membership degree) of the second layer; For x1=0.6, its membership degree as "high" is 0.8; For x2=0.8, its membership degree as "high" is 0.9; For x3=0.5, its membership degree to "middle" is 1.0; For x4=0.7, its membership degree as "high" is 0.6.

[0046] Step 2: Calculate the trigger strength w5 of the rule (third layer); w5=0.8×0.9×1.0×0.6=0.432; Physical meaning: In the current real-time state, the degree to which this rule is activated is only 43.2%.

[0047] Step 3: Calculate the local output f5 of the rule (third layer consequent); Assume the linear parameters of the rule are p=0.2, q=−0.1, r=−0.05, s=0.1, t=0.9; Substitute the actual values ​​of x1, x2, x3, and x4: f5=(0.2×0.6)+(−0.1×0.8)+(−0.05×0.5)+(0.1×0.7)+0.9=0.985; Physical implications: This rule strongly recommends maintaining the current speed, with only minor adjustments to 0.985.

[0048] Step 4: Summarize and output (Fifth layer); The fourth layer will sum up the w values ​​of all 81 rules (e.g., the total sum ∑w = 2.5), and then normalize it to obtain the result. =0.432 / 2.5=0.1728; The final correction coefficients output from the fifth layer: k v = ·f1+ ·f2+...+ ·f 81 .

[0049] The antecedent parameters (center value and width of the Gaussian function) and consequent parameters (coefficients p, q, r, s, and t of the first-order linear function) of the fuzzy neural network are continuously updated through a combination of offline training and online incremental learning. In the offline training phase, 500 sets of historical machining data (covering different materials, different tools, and different combinations of cutting parameters) are used for initial training using a hybrid learning algorithm (gradient descent for antecedent parameters and least squares for consequent parameters). In the online machining phase, after each workpiece is machined, the sensor data and machining quality data (surface roughness, dimensional deviation) collected during that machining process are used as new samples, and the network parameters are fine-tuned through backpropagation in a mini-batch manner (batch_size=32).

[0050] Furthermore, the offline training data of the fuzzy neural network is obtained in the following way: On the same type of CNC machine tool, using tools with different wear conditions (new tool, moderate wear, severe wear), orthogonal cutting experiments are carried out for three typical steel structural materials: Q235 steel, 45 steel and stainless steel, respectively, within the range of feed speed 100~500mm / min and spindle speed 500~10000rpm. A total of 500 sets of cutting data are collected. Each set of data includes vibration amplitude, average cutting force, temperature deviation and corresponding optimal feed speed and spindle speed.

[0051] Within each 50ms control cycle, the adaptive compensation module reads the latest sensor data from the shared memory circular buffer of the real-time monitoring module and inputs it into the ANFIS network to calculate k. v and k s Then, the parameters are adjusted according to the following decision logic: (a) When k v ≥0.9 and k s When the value is ≥0.95, the current processing status is determined to be normal, and the current processing parameters are maintained unchanged; (ii) When 0.8 ≤ k v <0.9 or 0.85≤k s When the value is less than 0.95, it is determined to be slight tool wear or a slight increase in material hardness. The feed rate or spindle speed is reduced by 0.95 times (which one is reduced depends on which correction factor is below the threshold). (iii) When 0.7 ≤ k v <0.8 or 0.75≤k s When the value is less than 0.85, it is considered as moderate tool wear, and the feed rate or spindle speed is reduced by 0.9 times. (iv) When k v <0.7 or k s When the value is less than 0.75 and this state lasts for more than three control cycles (i.e., more than 150ms), it is determined that the tool is severely worn. The adaptive compensation module sends a "tool replacement request" to the collaborative control module through internal communication events. After receiving the request, the collaborative control module jumps the state machine to the S9 pause state, the machining carriage 4 is raised to a safe height, the execution component stops operating, and the user interface displays the prompt "The tool is severely worn, please replace the tool" accompanied by a flashing alarm light. (v) When k v If the current cutting force is less than 60% of the rated value (2000N) (i.e., 1200N), and the current cutting parameters are deemed conservative, the feed rate is increased in increments of 1.05. After each increase, the k value is observed in the feedback of the next control cycle. v Value, if k vIf it is still greater than 1.1, continue to increase it until the feed rate reaches the maximum allowable feed rate limit of the tool or the cutting force reaches 80% of the rated value.

[0052] The aforementioned adaptive adjustments are performed continuously throughout the entire processing, ensuring that each process always operates under optimal processing parameters.

[0053] After the drilling (S3 state) and milling (S5 state) processes are completed, the machining quality self-inspection module automatically starts online inspection. This module integrates a radio probe on execution component 1 (42) and a line laser profile sensor on execution component 2 (43).

[0054] The inspection process after drilling is completed is as follows: The collaborative control module moves the execution component 42 to 2mm above the theoretical center position of the first hole to be tested. The machining quality self-inspection module drives the probe to descend along the Z-axis into the hole at a slow feed speed of 50mm / min. After the probe contacts the hole wall and triggers a signal, the coordinate value is recorded. The diameter and depth of each hole are measured three times and the arithmetic mean is taken. The system extracts no less than three holes from the completed drilling (if the total number of holes is less than three, all holes are inspected) for measurement, and compares the measured values ​​with the theoretical values ​​in the structured feature database. If the deviation is ≤0.05mm, it is judged as qualified, and the state machine continues to execute the subsequent process; if 0.05mm < deviation ≤0.15mm, the local rework process is automatically triggered: the machining quality self-inspection module calculates the coordinate compensation amount Δx=x 理论 -x 实测 and Δy=y 理论 -y 实测 The depth compensation value Δdepth = theoretical depth - measured depth is sent to the collaborative control module. The collaborative control module drives the execution component 42 to perform secondary drilling with the compensated coordinates and the compensated depth. After the secondary drilling is completed, the hole is remeasured. If it is still unqualified, compensation is performed again and a third drilling is performed. The maximum number of compensations for the same hole position is 3. If it is still unqualified after the third rework, an audible and visual alarm is issued and the hole is marked as "Rework failed, please intervene manually" on the user interface. If the deviation is >0.15mm, an alarm is directly triggered and it is marked as "Cannot be reworked".

[0055] The inspection process after the milling operation is completed is as follows: The collaborative control module moves the execution component 2 43 to the starting scanning position of the milling plane. The line laser profile sensor scans the entire machining plane at a speed of 100 mm / s along a direction parallel to the plane, collecting 5,000 to 10,000 point cloud data. The machining quality self-inspection module performs least squares fitting between the point cloud data and the theoretical plane to calculate the flatness error. If the flatness is ≤0.02 mm, it is judged as qualified; if the flatness is >0.02 mm, an error distribution map is generated based on the point cloud data, the spatial coordinate range of the out-of-tolerance area is identified, and a local compensation milling path is automatically generated (this path extends 5 mm outward from the out-of-tolerance area as the machining boundary). The execution component 2 43 is driven to mill the out-of-tolerance area again, and the scan and measurement are repeated after rework.

[0056] In addition, the machining quality self-inspection module also includes a statistical process control (SPC) unit, which automatically records the measurement data of each machining feature of each workpiece and generates a data set of 20 workpieces. Control chart (mean-range control chart). The SPC unit monitors the distribution of data points on the control chart in real time. When 7 consecutive data points are on the same side of the mean, or 3 consecutive data points exceed the 2σ limit (σ is the standard deviation of the current group of data), it is determined that a process deviation has occurred. The system issues a "process deviation warning" prompt on the user interface and suggests that the operator check the tool wear condition or the workpiece clamping stability.

[0057] After all machining and quality inspection are completed, the user interface displays "Machining Complete" and generates a quality inspection report. The report includes: total machining time, actual time for each process, maximum deviation value of each measurement feature and corresponding hole / face position, flatness error value, number of reworks and rework results, and statistical summary of all sensor data (maximum value, minimum value, average value). Simultaneously, controller 5 displays all parameters of this machining process (including the optimized process sequence, actual feed rate and spindle speed of each process, and k-cycle of each control cycle of the adaptive compensation module). v and k s The data (including values) and all sensor data are automatically saved to the internal 64GB eMMC storage medium. The file name is named according to the processing time and workpiece number, for subsequent offline analysis and process optimization.

[0058] Through the coordinated work of the above modules, the CNC machine tool in this embodiment achieves fully automated intelligent control from the import of the 3D model to the completion of the finished product processing, without the need for manual programming or real-time intervention by operators.

[0059] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A CNC machine tool for composite steel structure processing, comprising a base frame (1), characterized in that: A Y-axis slide saddle (2) is provided above the base frame (1), and the Y-axis slide saddle (2) is equipped with a Y-axis moving component (21) for driving the Y-axis slide saddle (2) to adjust along the Y-axis direction; A support platform (23) is slidably disposed above the Y-axis slide saddle (2), and the support platform (23) is equipped with an X-axis moving component (22) for driving the support platform (23) to adjust along the X-axis direction; Clamping assemblies (24) are installed on both sides above the support platform (23) for clamping the workpiece to be processed. The base frame (1) is equipped with a gantry frame (3) on both sides. A processing slide (4) is provided on one side of the gantry frame (3). The processing slide (4) is equipped with a Z-axis lifting assembly (41) for driving the processing slide (4) to adjust along the Z-axis direction. The Z-axis lifting assembly (41) is installed on the gantry frame (3). The processing carriage (4) has an execution component one (42) installed on the top of one side, an execution component two (43) installed on the right side, and an execution component three (44) installed on the left side. A control component (5) is installed on one side of the gantry (3), the control component (5) comprising: The model import and feature recognition module is used to receive the 3D model file of the workpiece to be processed and automatically extract hole features, planar features and curved surface features; The processing sequence planning module, connected to the feature recognition module, is used to generate a processing sequence of drilling, milling, and grinding based on the extracted features, and to plan the calling order and processing parameters of each execution component. The path generation and simulation module is used to generate the tool motion trajectory for each machining process, perform interference and collision detection, and output a collision-free machining path. The real-time monitoring module is used to collect signals from vibration sensors, cutting force sensors, and temperature sensors during the machining process; The adaptive compensation module dynamically adjusts the feed rate, spindle speed, and depth of cut based on data collected by the real-time monitoring module using a fuzzy neural network model. The collaborative control module is used to synchronously control the X-axis movement component (22), the Y-axis movement component (21), the Z-axis lifting component (41), and the execution component one (42), the execution component two (43), and the execution component three (44).

2. The CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The model import and feature recognition module adopts a dual-channel parallel feature recognition architecture: The first channel is a processing feature classification channel based on graph neural network. Each face of the 3D model is treated as a node in the graph, and the adjacency relationship between faces is treated as an edge. An attribute adjacency graph is constructed. The attribute adjacency graph is processed by a spectral domain-spatial domain hybrid graph convolutional network to identify through holes, threaded bottom holes, countersunk holes, waist-shaped holes, right-angled planes and arc surfaces, and output the spatial pose and geometric parameters of each feature. The second channel is a fine feature segmentation channel based on point cloud deep learning. It uniformly samples the surface of the 3D model into point cloud data, performs semantic segmentation on the point cloud through the PointNet++ network, and classifies each point into regions, including drilling regions, milling regions, grinding regions and non-processing regions. The recognition results from the two channels are merged by a feature fusion layer to generate a structured feature database containing feature ID, feature type, spatial pose, geometric parameters, tolerance level, and surface roughness requirements.

3. The CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The processing sequence planning module includes a multi-objective optimizer based on the NSGA-II non-dominated sorting genetic algorithm. The decision variables of the multi-objective optimizer are the process arrangement vector and the continuous processing parameters of each process. The optimization objective is to minimize the total processing time T. total and minimize tool wear W total ; The total processing time T total =Σ(t move，i +t machine，i +t toolchange，i ), where t move，i The idle travel time between adjacent workstations, t machine，i For actual cutting time, t toolchange，i This is the tool change time; The tool wear W total =Σ(k material ×k tool ×L i ), where k material For the wear coefficient of the workpiece material, k tool For the wear coefficient of the tool material, L i Let be the cutting path length of the i-th process; The constraints of the multi-objective optimizer include: available rotational speed range constraints for each execution component, maximum allowable feed rate constraints, maximum allowable cutting force constraints, spindle power constraints, process sequence constraints between operations, and geometric reachability constraints between the execution component and the workpiece; The multi-objective optimizer outputs a Pareto front solution set after population iterative evolution, and the inflection point solution is selected by the operator or automatically selected by the system as the final optimization processing sequence.

4. The CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The path generation and simulation module incorporates a gantry machine tool forward kinematics model and an S-shaped acceleration / deceleration planner, converting the tool path into motion commands for each servo axis, and employs a three-layer progressive collision detection strategy. The first layer is a coarse axial bounding box (AABB) test, which encloses each object in an axial bounding box and tests the intersection. The second layer involves detection within the directional bounding box (OBB). For objects that intersect within the AABB, the directional bounding box is used to perform more accurate intersection detection. The third layer is based on the Separated Axis Theorem (SAT) for fine detection of triangular facets. For object pairs where OBB still intersects, the triangular mesh model is extracted and interference is detected on a triangular facet-by-triangular facet basis. If interference is detected, the path generation and simulation module automatically triggers obstacle avoidance path replanning: randomly sample collision-free poses in the configuration space near the interference point, and use RRT* to quickly explore the random tree star topology algorithm to connect the blocked trajectories. If the number of retries exceeds the preset threshold and a collision-free path still cannot be found, the process is marked as requiring manual intervention and a prompt is issued.

5. A CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The real-time monitoring module includes a triaxial accelerometer, a strain-type triaxial cutting force sensor, a thermocouple temperature sensor, a spindle current Hall effect sensor, an eddy current displacement sensor, an FPGA coprocessor, and an ARM processor. The signals from the vibration sensor and the cutting force sensor are digitized by a 24-bit Δ-Σ ADC and then preprocessed by an FIR digital filter by the FPGA coprocessor. The FIR digital filter includes a fourth-order Butterworth bandpass filter for the vibration signal and a fourth-order Butterworth lowpass filter for the cutting force signal. The preprocessed data is written to the shared memory ring buffer on the ARM processor side via DMA direct memory access. Each data packet is accompanied by a hardware timestamp for the adaptive compensation module to read.

6. A CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The fuzzy neural network model in the adaptive compensation module is a five-layer adaptive neurofuzzy inference system (ANFIS) architecture. The first layer is the input layer, which receives four input variables: the normalized effective value of vibration amplitude, the average value of cutting force, the temperature deviation, and the current feed rate. The second layer is the membership function layer, where each input variable corresponds to three Gaussian membership functions, representing the low, medium and high levels of state respectively. The third layer is the fuzzy rule layer, which calculates the trigger strength of each rule by multiplying the membership degrees of each input. The consequent of the rule is a first-order linear function. The fourth layer is the normalization layer, which calculates the normalized trigger strength of each rule; The fifth layer is the output layer, which outputs the feed rate correction coefficient k. v and spindle speed correction factor k s ; The antecedent and consequent parameters of the fuzzy neural network model are continuously updated through a combination of offline training and online incremental learning. During the offline training phase, a hybrid learning algorithm is used for initial training with historical processed data; During the online processing phase, after each workpiece is processed, the currently collected sensor data and processing quality data are used as new samples to fine-tune the network parameters through backpropagation in a small batch manner.

7. A CNC machine tool for composite machining of steel structures according to claim 6, characterized in that, The adaptive compensation module adjusts parameters according to the following decision logic: When k v ≥0.9 and k s When the value is ≥0.95, maintain the current processing parameters; When 0.8≤k v <0.9 or 0.85≤k s When the value is less than 0.95, reduce the feed rate or spindle speed by 0.95 times. When 0.7≤k v <0.8 or 0.75≤k s When the value is less than 0.85, reduce the feed rate or spindle speed by 0.9 times. When k v <0.7 or k s If the value is less than 0.75 and persists for more than three control cycles, a tool change warning will be triggered and machining will be paused. When k v When the feed rate is greater than 1.1 and the current cutting force is less than 60% of the rated value, the feed rate is increased by 1.05 times until the upper limit is reached.

8. A CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The collaborative control module has a built-in hierarchical finite state machine. The top-level master state machine includes the following states: S0 Standby, S1 Clamping Detection, S2 Zeroing, S3 Drilling Process, S4 First Tool Change Position Adjustment, S5 Milling Process, S6 Second Tool Change Position Adjustment, S7 Grinding Process, S8 Machining Completed, S9 Pause and S10 Emergency Stop. The state machine automatically jumps according to the process list output by the processing sequence planning module, and the state transition is driven by external events, internal events and communication events; The main states of the drilling, milling and grinding processes are each equipped with a sub-process state machine. The sub-process state machine includes positioning, rapid advance, feed, pause, tool retraction and position change sub-states. Each sub-state corresponds to a set of axis motion commands and I / O operations. All state transitions are recorded in the system log with timestamps.

9. A CNC machine tool for composite machining of steel structures according to claim 1, characterized in that, The control component (5) also includes a processing quality self-inspection module, which integrates a contact radio probe on the execution component one (42) and a line laser profile sensor on the execution component two (43); The contact-type wireless probe is used to measure the diameter and depth of no less than three drilled holes after the drilling process is completed. If the measurement deviation exceeds the first preset threshold, the local rework process is automatically triggered: the control execution component (42) returns to the out-of-tolerance hole position, calculates the coordinate compensation amount and depth compensation amount according to the deviation value and performs secondary drilling. The maximum number of compensations for the same hole position is three. An alarm is issued after the maximum number of compensations is exceeded. The line laser profile sensor is used to scan the machining plane after the milling process to generate point cloud data, compare it with the theoretical plane to calculate the flatness error, and if the flatness error exceeds the second preset threshold, a local compensation path is generated according to the error distribution map to drive the execution component two (43) to mill the out-of-tolerance area again. The processing quality self-inspection module also includes a statistical process control unit, which automatically records the measurement data of each workpiece and generates a mean-range control chart. When seven consecutive data points are on the same side of the mean or three consecutive data points exceed the 2σ limit, a process deviation warning is issued.

Images