CN-122018433-A - System and method for self-adaptive optimization of machining process and recognition of abnormal working condition

Abstract

The invention provides a processing self-adaptive optimization and abnormal working condition identification system and method, which adopt a layered decoupling architecture, a front-end display layer is used as an interactive interface, display contents of the front-end display layer are acquired from a core algorithm layer through a data transfer and storage support layer, the core algorithm layer adopts a real-time and non-real-time service separation architecture, the two are decoupled to ensure that real-time processing is not affected, a hardware interaction layer is used for realizing bidirectional data interaction with a numerical control system and a sensor of a machine tool through an adaptive device and covering data acquisition and feedback control, the data transfer and storage support layer adopts a Redis publishing and subscribing mechanism, a plurality of fixed channels are divided to realize data transmission decoupling, and the system acquires processing full-dimension data through three-source data acquisition. The method has the advantages of decoupling of each layer and high data transmission efficiency, effectively improves monitoring instantaneity, optimizing accuracy and anomaly identification reliability in the processing process, and reduces loss caused by working condition anomalies.

Inventors

• ZHENG ZUJIE
• Quan Zhongxin
• SHEN CAIXIA
• SHEN FANGFANG
• JIAO BINBIN
• PENG CHENYUAN
• CHEN XUEFEN
• Gong Youyong
• SHEN YI
• YU HAIYANG

Assignees

• 上海航天精密机械研究所

Dates

Publication Date

20260512

Application Date

20260413

Claims (10)

1. 1. A processing self-adaptive optimization and abnormal working condition identification system is characterized by adopting a layered decoupling architecture, wherein the layered decoupling architecture comprises a front-end display layer, a core algorithm layer, a hardware interaction layer and a data transfer and storage support layer; The front-end display layer is a display and interaction interface of the identification system, and display content of the front-end display layer is obtained from the core algorithm layer through the data transfer and storage support layer; The core algorithm layer adopts a separated architecture comprising real-time service and non-real-time service, wherein the real-time service is used for completing real-time processing of data, the non-real-time service is used for executing time-consuming tasks, and the real-time service and the non-real-time service are mutually decoupled; The hardware interaction layer respectively performs bidirectional data interaction with a machine tool numerical control system and a sensor through a numerical control system communication adapter and an external sensor acquisition card, wherein the bidirectional data interaction comprises data acquisition of a machining process and feedback control of the machining process; the data transfer and storage support layer adopts a Redis publishing and subscribing mechanism, and is divided into a plurality of fixed data transmission channels for realizing data transmission decoupling; The identification system acquires full-dimension data in the processing process by adopting a three-source data acquisition mode.
2. 2. The process adaptive optimization and anomaly identification system of claim 1, wherein the real-time service uses 50ms as a fixed operation period; The fixed data transmission channels comprise a high-frequency real-time channel, a low-frequency real-time channel, a command issuing channel and a command feedback channel, wherein the data transmission frequency of the high-frequency real-time channel is more than or equal to 20Hz and is used for continuously updating and transmitting data, the data transmission frequency of the low-frequency real-time channel is less than 20Hz and is used for continuously updating and transmitting data, the command issuing channel is used for issuing a front-end interactive interface request command, and the command feedback channel is used for returning a rear-end processing result; And/or the data acquired by adopting the three-source data acquisition mode comprises machine tool internal data acquired from a machine tool numerical control system through an HNC adapter or an NC-link protocol, processing external data acquired from an external vibration sensor through a sensor acquisition card and NC code execution data in the processing process.
3. 3. The system for processing adaptive optimization and abnormal condition identification according to claim 1, wherein the core algorithm layer comprises a multi-dimensional monitoring visualization module, an adaptive optimization control module and an abnormal condition identification prediction module; The multi-dimensional monitoring visualization module constructs a space-time coupled digital twin monitoring environment, dynamically correlates a time domain monitoring signal of a machining process with a three-dimensional space position, and realizes workshop site multi-dimensional visual presentation of the numerical control machining process through color mapping and a state machine; The self-adaptive optimization control module dynamically calculates a feeding rate optimization value by adopting a fuzzy control algorithm based on the real-time load data and an optimization parameter database constructed by offline analysis, and writes and executes the optimization parameters in real time through a numerical control system interface; the abnormal working condition identification prediction module establishes a learning and monitoring mechanism based on a main shaft power signal, performs sliding window statistics and boundary fusion on a plurality of historical qualified machining records of the same numerical control machining task to generate an envelope curve model, monitors whether the vibration intensity exceeds an alarm boundary, a wear boundary or a load lower limit in real time in the machining process, and further realizes identification and early warning of abnormal working conditions including tipping, tool bumping and clamping looseness.
4. 4. The processing self-adaptive optimization and abnormal working condition identification system according to claim 3, wherein the multi-dimensional monitoring and visualization module comprises a space-time coupling monitoring unit, a spatial position monitoring unit, a state machine and UI interaction unit and a UI interface partition design unit; the space-time coupling monitoring unit adopts ECharts dynamic line diagram components to display the main shaft load, the three-way vibration intensity, the synthesized vibration intensity and the time signals in real time at a preset refresh rate, and supports multi-data source switching and data item screening; Front-end and back-end real-time communication is realized through a WebSocket pushing mechanism, a history mean curve and an alarm threshold line are superimposed in a chart, and the deviation amount of a real-time state and a history reference is calculated according to the following steps: Wherein, the For the signal value at the current moment, For the signal mean of the corresponding positions of the history and the code segments, Is the standard deviation when Triggering an abnormality flag when the number is greater than 3; The space position monitoring unit constructs a three-dimensional rendering scene based on a Python VTK engine and sets CuttingTool classes and Part classes, wherein the CuttingTool classes comprise cutter type, cutter diameter, cutter edge length and cutter overhang quantity attributes, and the Part classes comprise workpiece names, workpiece materials, a workpiece STL model vertex array and workpiece normal vector attributes; The space position monitoring unit reads a tool and workpiece model in an STL format, builds an octree space index to accelerate collision detection, and realizes real-time visualization of a tool motion trail through a 4×4 homogeneous coordinate transformation matrix, wherein the homogeneous transformation matrix is as follows: Wherein, the For a 3 x 3 rotation matrix, And simultaneously adopting the color shade mapping to process the data size at the corresponding position, wherein the color mapping function is as follows: Wherein, the Is the position At which the vibration intensity or load value is calculated, 、 Global minimum and maximum of the current presentation data, Is an iridescent look-up table function; The state machine and UI interaction unit is used for constructing a finite state machine model, defining a stable state and a state transition event, wherein the stable state comprises S0 standby, S1 start, S2 processing and S3 end, and the state transition event is used for updating the color of a UI top column state indicator lamp in real time, namely, standby blue, yellow flashing, green processing and gray ending, triggering corresponding business logic when the state is changed, and automatically loading an optimized parameter library when the state is changed into a processing state; the UI interface partition design unit adopts Flex elastic layout to divide the interface into a plurality of areas, including a top tool bar, a left side status bar, a central main view area, a bottom NC code bar and a bottom status bar; The top toolbar is used for realizing space-time monitoring view switching, abnormal monitoring view switching, self-adaptive function switches and system setting, the left side status bar is used for displaying machine tool status word description, current program name, operator information and current alarm list, the central main view area is used for displaying three-dimensional rendering canvas and real-time graphs, the bottom NC code bar is used for highlighting current execution code lines in real time and supporting clicking positioning to a corresponding position of a three-dimensional view, and the bottom status bar is used for displaying database connection status, activation status of each function module, system time and network delay.
5. 5. The system for processing adaptive optimization and abnormal condition identification according to claim 3, wherein the adaptive optimization control module comprises an optimization parameter database construction unit, an intelligent control algorithm unit and a parameter execution interface unit; The optimization parameter database construction unit extracts an empty load P 1 and a target load P 2 according to the combination of each program name and a cutter number by analyzing historical processing data offline; the no-load P 1 is defined as satisfying P 1 ≥P min +0.1×(P max P min ), wherein the load value is smaller than the minimum power value of which the continuous data duration time of P 1 is not smaller than 5s, and transient fluctuation data points of cutting fluid at the moment of opening or closing are removed, P max refers to the maximum value of the collected load power in the processing process corresponding to the same program name and cutter number combination, and P min refers to the minimum value of the collected load power in the processing process corresponding to the same program name and cutter number combination; The target load P 2 is defined as a minimum power value with the data point of which the load value is smaller than P 2 in the machining stable section and the data point of which the data point is not smaller than 95 percent, and is used for representing the typical load level in the machining stable section; The intelligent control algorithm unit adopts a two-dimensional fuzzy PID control algorithm to control the error E (t) =P between the real-time load P (t) and the target load P 2 2 P (t) and error differential Ec (t) =de/dt are input amounts, and are mapped to the fuzzy domain through quantization factor K e 、K ec : E * =round(K e E),K e =6/P 2 ×0.5 Ec * =round(K ec Ec),K ec =6/P 2 ×0.1 The input fuzzy subsets of E * and Ec * are { NB, NM, NS, ZO, PS, PM, PL }, the domain of error E (t) is { -6, -4, -2,0,2,4,6}, the domain of error differential Ec is { -6, -4, -2,0,2,4,6}, the output is the feed rate increment DeltaU, the fuzzy subsets are consistent with the input, the domain of domain is { -3, -2, -1,0,1,2,3}, and the membership function adopts triangle distribution; Constructing a fuzzy rule table, performing fuzzy reasoning by adopting a Mamdani synthetic reasoning method, and performing fuzzy solution by a weighted average method: Wherein mu i is the activation degree of the ith rule, C i is the central value of the output fuzzy set of the corresponding rule, and the final output control quantity DeltaU=Ku Δu, ku is a scale factor, an actual feeding rate optimization value F opt (t+1)=F opt (t) +Δu is defined within a 30% -150% interval, F opt (t) refers to an actual feeding rate optimization value at the current time t, and a vibration feedback correction term is introduced simultaneously: wherein V (t) is the real-time vibration intensity, For the target vibration reference, eta is the vibration suppression coefficient, when the vibration intensity exceeds 1.5 times Executing active deceleration control; The parameter execution interface unit is used for writing R parameters into a numerical control system through an OPC UA method node and calling a synchronous action instruction to realize multiplying power coverage activation, and a double-channel redundancy writing mechanism is adopted, and the multiplying power effective success rate is not lower than 99.9% through double writing of a PLC data block channel 1 multiplying power DB21 and a shaft multiplying power DB 31; And the final feeding multiplying power F final =F opt ×F panel /100%, wherein F panel is a control panel knob set value, and fusion control of the optimized multiplying power and manual setting is realized.
6. 6. The processing self-adaptive optimization and abnormal working condition identification system according to claim 3, wherein the abnormal working condition identification prediction module comprises a learning model construction unit, a collision detection unit and an abnormal alarm and backtracking unit; The learning model construction unit is used for analyzing the time domain characteristic change of the spindle power signal so as to realize real-time judgment of abnormal working conditions including tipping, tool collision and clamping looseness; The learning model construction unit supports training corresponding independent learning models for different numerical control programs respectively, and each model is associated with a program name and a cutter number and is used for monitoring and identifying abnormality in the repeated processing process of the same subsequent program; the generation process of the learning model comprises the following steps: Selecting a sample, namely selecting a single or a plurality of qualified processing records with the same cutter number under the same program name as a training sample set; Setting the length W of a time window to be 0.5s, setting the deviation value delta alarm of an alarm boundary to be 0.2g, setting the deviation value delta wear of a wear boundary to be 0.1g, and setting the deviation value delta low of a load lower limit to be 0.05 times of a load fluctuation range; A time axis alignment step of performing time stamp homogenization treatment on each processing record in the training sample set by adopting a dynamic time warping algorithm to align time axes among a plurality of processing records; The early warning boundary calculation step comprises the steps of traversing the whole processing process by using the length W of a time window for each processing record, calculating the statistical characteristics of monitoring signals in a window at the position t i of each time window, and generating an upper early warning boundary sequence and a lower early warning boundary sequence of the record; The upper early warning boundary point is defined as: the lower early warning boundary point is defined as: Wherein, the For the sequence of spindle power signals within the window, As the maximum value of the power in the window, As the minimum value of the power in the window, As an average value of the power in the window, Is a boundary coefficient; The envelope curve fusion step, namely fusing upper and lower early warning boundaries generated by each record at the ith time window position t i aiming at a sample set containing K learning records to obtain a final envelope curve model: Upper envelope curve: , The lower envelope curve: , Wherein, the 、 The k-th record is recorded on the upper and lower pre-warning boundaries of the i-th window, 、 A value of the paranoid set for the user, The value of (a) corresponds to delta alarm , Corresponding to delta wear or delta low ; dynamic boundary generation based on the fused envelope curve 、 Three boundary curves for real-time monitoring are generated: Alarm boundary B alarm (t)=U env (t)+Δ alarm ×σ U (t) Wear boundary B wear (t)=U env (t) Δ wear ×σ U (t) Lower load limit B low (t)=L env (t) Δ low ×σ L (t) Wherein sigma U (t)、σ L (t) is the standard deviation of each learning record at the corresponding time t of the upper boundary and the lower boundary and is used for representing the data fluctuation degree of the sample set; During the machining process, the collision detection unit is used for monitoring the vibration intensity signal V (t) in real time and comparing the vibration intensity signal V (t) with the three boundary curves so as to judge the state of the cutter: when V (t) > B alarm (t), judging that the vehicle is seriously abnormal, including tipping or bumping, and triggering an emergency stop instruction; when B wear (t)<V(t)≤B alarm is carried out, judging that the cutter is worn or slightly abnormal, and triggering a secondary alarm; When V (t) < B low (t), judging that clamping looseness or empty cutting is abnormal, and triggering primary early warning; The learning model construction unit supports model version management and iterative optimization, and specifically comprises the following steps: the model evaluation step comprises the steps of selecting an optimal model by comparing prediction errors of a plurality of groups of learning models on the same verification set, wherein the prediction errors comprise Root Mean Square Error (RMSE) and average absolute percentage error (MAPE); the root mean square error RMSE calculation formula is as follows: The average absolute percentage error MAPE calculation formula is as follows: And an incremental updating step, namely periodically utilizing newly accumulated qualified processing data to perform incremental learning updating on the selected optimal model so as to adapt to the change of the production working condition.
7. 7. The adaptive optimization and abnormal working condition identification system for machining processes according to claim 6, wherein the collision detection unit is used for monitoring the vibration intensity of the turning/milling channel in real time, setting a collision judgment threshold value through a preset formula, and triggering a machine tool stop signal when the vibration intensity monitored in real time exceeds the collision judgment threshold value; the calculation formula of the collision judgment threshold value is as follows: T collision =V ref ×(1+α)+β σ ref wherein V ref is the reference vibration intensity, alpha is the reference percentage coefficient, beta is the dynamic margin coefficient, and sigma ref is the reference vibration standard deviation; The reference vibration intensity is obtained through the learning model construction unit through the learning qualified machining record, the reference vibration standard deviation corresponds to the reference vibration intensity, and the reference vibration intensity is obtained through the learning model construction unit through the learning qualified machining record synchronously.
8. 8. The adaptive process optimization and anomaly identification system of claim 6, wherein the anomaly alarm and backtracking unit triggers a classification alarm and shutdown command based on a comparison of vibration intensity and a predetermined boundary: When the vibration intensity exceeds the abrasion boundary or the lower load limit, the abnormal alarming and backtracking unit triggers the grading alarming; when the vibration intensity exceeds the alarm boundary, the abnormal alarm and backtracking unit triggers a machine tool stopping instruction; the abnormal alarm and backtracking unit supports an abnormal event backtracking function, and can be used for calling and analyzing parameters, waveforms and state records related to the alarm in the current and historical processing processes.
9. 9. A method for processing adaptive optimization and abnormal condition identification, characterized in that the method is realized based on the processing adaptive optimization and abnormal condition identification system of any one of claims 1 to 8, and comprises the following steps: The system initialization step is that a layered decoupling architecture of the system is started, a numerical control machine tool configuration parameter, a sensor acquisition parameter and a learning model library stored in the system are loaded, initialization deployment of each functional module is completed, and normal linkage of a front-end display layer, a core algorithm layer, a hardware interaction layer and a data transfer and storage support layer is ensured; The learning model construction step, namely selecting qualified processing records corresponding to the history of a preset numerical control processing program as training samples, and generating an envelope curve model by adopting a sliding window statistics and boundary fusion algorithm through a learning model construction unit of an abnormal working condition identification prediction module of a core algorithm layer, wherein the envelope curve model is stored in association with the numerical control processing program and a corresponding cutter number; Acquiring full-dimensional data of a machining process in a three-source data acquisition mode of a system to form three-source data, wherein the three-source data comprises internal data acquired from a machine tool numerical control system through an HNC adapter or an NC-link protocol, external data acquired from an external vibration sensor through a sensor acquisition card and NC code execution data, and the acquired data is distributed to corresponding functional modules through a plurality of fixed channels of a data transfer and storage support layer; A multi-dimensional monitoring visualization step of driving three-dimensional digital twin monitoring scene updating and real-time dynamic curve drawing based on real-time data distributed in the real-time processing monitoring step by a multi-dimensional monitoring visualization module of a core algorithm layer to realize multi-dimensional visualization presentation of a processing process; The self-adaptive optimization control step is that the self-adaptive optimization control module of the core algorithm layer is used for calculating the feeding multiplying power optimization value in real time by combining the real-time load data with the optimization parameter database based on the fuzzy control algorithm, and writing the feeding multiplying power optimization value into the numerical control system for execution through the parameter execution interface unit; An abnormal working condition identification step, namely acquiring a main shaft power signal and a vibration signal through a collision detection unit and a learning model construction unit of an abnormal working condition identification prediction module of a core algorithm layer, comparing the real-time vibration signal with an alarm boundary, a wear boundary and a load lower limit corresponding to an envelope curve model generated in the learning model construction step in real time, judging a processing abnormal state, and triggering corresponding grading alarm, shutdown or early warning treatment through an abnormal alarm and backtracking unit of the abnormal working condition identification prediction module of the core algorithm layer; And a data storage and model iteration step, namely storing the processing process data acquired in the real-time processing monitoring step, the optimization parameter data in the self-adaptive optimization control step and the abnormal identification data in the abnormal working condition identification step into a database of a data transfer and storage support layer, and periodically utilizing the newly accumulated qualified processing data to perform incremental training on a learning model through a learning model construction unit so as to realize model iteration optimization.
10. 10. The method for adaptive optimization and anomaly identification of a machining process of claim 9, wherein in the step of real-time machining monitoring, the three-source data comprises: The first source data comprises that a spindle load, coordinates of each axis and a feeding multiplying power are collected from a machine tool numerical control system through an HNC adapter or an NC-link protocol; the second source data is that three-way vibration signals of the turning/milling channel are collected from an external vibration sensor through a sensor collection card; the third source data is to obtain the current execution line number and the code content through NC code analysis; The data transfer and storage support layer is divided into four fixed data transmission channels by adopting a Redis publishing and subscribing mechanism, and the four fixed data transmission channels are respectively: the high-frequency real-time channel transmits monitoring data and data required by digital twin scene updating in the processing process at a data transmission frequency of more than or equal to 20 Hz; the low-frequency real-time channel transmits data generated by the model and requested by the interface at a data transmission frequency of less than 20 Hz; The command issuing channel is used for receiving a request command issued by the front-end display layer and forwarding the request command to the corresponding function module at the rear end; and the command feedback channel is used for receiving the processing results of the functional modules at the back end and retransmitting the processing results to the front-end display layer so as to realize front-end and back-end data interaction.

Description

System and method for self-adaptive optimization of machining process and recognition of abnormal working condition Technical Field The invention relates to the technical field of mechanical processing and industrial automation, in particular to a processing process self-adaptive optimization and abnormal working condition identification system and method, especially a processing process self-adaptive optimization and abnormal working condition identification system and method driven by multi-dimensional real-time monitoring, especially a processing process data acquisition, multi-dimensional monitoring visualization, self-adaptive process parameter optimization and abnormal working condition identification system, which is suitable for numerical control processing scenes of aerospace complex structural members. Background In the numerical control machining process of a space complex structural member, the space complex structural member has the characteristics of complex structure (mostly thin wall, deep cavity and special-shaped curved surface structure), extremely high machining precision requirement (directly influencing the assembly precision and the operation reliability of space equipment), most difficult-to-machine materials (high hardness, strong toughness, large cutting resistance) such as high-temperature alloy and composite material, complicated machining process, small production batch and high customization degree, so that the machining process is influenced by various factors, and the following prominent technical pain points are faced: a. The monitoring dimension of the machining process is single, the traditional monitoring method of the numerical control machining process only focuses on time domain signals such as spindle load, vibration or space position information, but the coupling correlation between the two is poor, machining data corresponding to the cutting position of a workpiece cannot be accurately positioned, the data are difficult to effectively and directly analyze, and the machining state judgment is incomplete; b. The process parameter optimization depends on manual experience or off-line setting, so that the variable cutting working conditions are difficult to adapt, the problems of cutter load surge, tipping and the like easily occur at the difficult-to-machine part, the cutting efficiency is greatly reduced due to conservative parameter setting, and the performances of a machine tool and a cutter cannot be fully exerted; c. the abnormal working condition identification mostly adopts a single signal static threshold method, only judges aiming at single parameters such as vibration, load and the like, has high false alarm rate, lacks prediction capability for complex anomalies such as cutter abrasion, collision, clamping looseness and the like, and is difficult to realize early warning and tracing; d. The processing data volume is large, the types are heterogeneous, the real-time transmission and the processing are easy to be blocked, the cooperative coupling degree of multiple modules (monitoring, optimizing and alarming) is low, and the overall response speed of the system is influenced. In the prior art, processing monitoring, parameter optimization and anomaly identification are mostly independent systems, and lack of an integrated scheme, so that the processing requirements of high precision, high efficiency and high reliability of a space complex structural member cannot be met. Disclosure of Invention Aiming at the defects in the prior art, the invention aims to provide a system and a method for processing process self-adaptive optimization and abnormal working condition identification. According to the processing self-adaptive optimization and abnormal working condition identification system provided by the invention, a layered decoupling architecture is adopted, wherein the layered decoupling architecture comprises a front-end display layer, a core algorithm layer, a hardware interaction layer and a data transfer and storage support layer; The front-end display layer is a display and interaction interface of the identification system, and display content of the front-end display layer is obtained from the core algorithm layer through the data transfer and storage support layer; The core algorithm layer adopts a separated architecture comprising real-time service and non-real-time service, wherein the real-time service is used for completing real-time processing of data, the non-real-time service is used for executing time-consuming tasks, and the real-time service and the non-real-time service are mutually decoupled; The hardware interaction layer respectively performs bidirectional data interaction with a machine tool numerical control system and a sensor through a numerical control system communication adapter and an external sensor acquisition card, wherein the bidirectional data interaction comprises data acquisition of a machining process and feedback control