Self-sensing and intelligent early warning method for stress of numerically controlled machining aluminum alloy shell part

Abstract

The application provides a self-sensing and intelligent early warning method for stress of numerically controlled machining aluminum alloy shell parts, and particularly relates to the online intelligent monitoring and early warning technical field of numerically controlled machining. The method forms a conformal self-sensing array by distributing and embedding micro piezoelectric ceramic units on the surface of the aluminum alloy shell part, generates a piezoelectric charge signal, sends a self-sensing signal containing stress information after radio frequency identification backscattering modulation, receives the signal and synchronously acquires real-time machining parameters of the numerically controlled machining system, generates a real-time fusion feature vector through fusion processing of a self-adaptive weight distribution model, inputs a pre-trained pulse neural network model, outputs a stress risk probability value, synchronizes to a digital twin model for three-dimensional thermal map visualization display, and generates an execution hierarchical early warning or machining control instruction according to the stress risk probability value and a dynamic early warning threshold. The application realizes real-time, passive and distributed sensing and dynamic risk assessment of the machining stress state, and improves the timeliness of early warning and the self-controllability of machining.

Description

Technical Field

[0001] This invention relates to the field of online intelligent monitoring and early warning technology for CNC machining, and more specifically, to a method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts. Background Technology

[0002] With the rapid development of high-end equipment manufacturing industries such as aerospace and new energy vehicles, the demand for lightweight, high-precision aluminum alloy shell parts is increasing. CNC machining technology plays a core role in the manufacturing of such parts. However, during CNC milling, drilling, and other machining processes, residual stress is easily generated and accumulated inside the parts due to the effects of cutting forces, thermal loads, and uneven material removal. This can lead to problems such as machining deformation, dimensional deviations, and even cracks, severely restricting the machining accuracy, consistency, and service reliability of the parts.

[0003] Currently, the industry mainly uses two types of technologies to address the aforementioned stress problems: one is pre-processing prediction based on finite element simulation, which simulates stress distribution by establishing a material constitutive model and process parameters. However, it involves a large amount of computation, relies on precise boundary conditions, and is difficult to reflect the stress evolution during dynamic processing in real time. The other is online monitoring based on external sensors, such as attaching strain gauges to the surface of the part or integrating fiber optic sensors. This method can acquire real-time data, but the sensors usually require external power supply and wiring, which can easily interfere with the processing environment. Furthermore, the monitoring points are limited, making it difficult to achieve distributed sensing of the entire part itself. Moreover, it lacks the ability to intelligently evaluate and warn of the coupling relationship between processing parameters and stress state.

[0004] In summary, existing technologies rely on active or external monitoring methods, which makes it difficult to achieve stress self-sensing of aluminum alloy shell parts without interfering with the machining process, at low cost and covering the entire area. This specific deficiency is a problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The main objective of this invention is to provide a self-sensing and intelligent early warning method for stress in CNC-machined aluminum alloy shell parts. This method addresses the specific shortcomings of existing technologies that rely on active or external monitoring methods, which prevent the non-interfering, low-cost, and fully covered stress self-sensing of aluminum alloy shell parts during CNC machining. The invention achieves real-time, passive, and distributed sensing of machining stress state and dynamic risk probability assessment, significantly improving the timeliness of early warning and the autonomous controllability of the machining process.

[0006] To achieve the above objectives, the present invention provides a method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts.

[0007] This invention provides a method for self-sensing and intelligent early warning of stress in CNC-machined aluminum alloy shell parts, the method comprising:

[0008] Multiple micro piezoelectric ceramic units are distributedly embedded on the surface of the aluminum alloy shell part to form a self-sensing array conforming to the surface of the aluminum alloy shell part. The micro piezoelectric ceramic units are used to generate piezoelectric charge signals based on the stress applied during CNC machining.

[0009] The piezoelectric charge signal is used as a modulation source, and the modulation source is modulated by radio frequency identification backscattering to obtain a reflected radio frequency signal. The reflected radio frequency signal is then sent as a self-sensing signal containing stress information.

[0010] The system receives the self-sensing signal and synchronously acquires the real-time machining parameters of the CNC machining system; wherein the CNC machining system is used to indicate the CNC system controlling the CNC machining process.

[0011] The self-sensing signal and the real-time machining parameters are fused using an adaptive weight allocation model to generate a real-time fused feature vector. The adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process.

[0012] The real-time fused feature vector is input into a pre-trained spiking neural network model. The spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value that characterizes the current stress risk level of the aluminum alloy shell part.

[0013] A digital twin model of the aluminum alloy shell part is constructed, and the stress risk probability value and the real-time processing parameters are synchronized to the digital twin model. Based on the stress risk probability value, a three-dimensional thermal map is visualized on the three-dimensional model of the aluminum alloy shell part corresponding to the digital twin model.

[0014] Based on the stress risk probability value and dynamic early warning threshold, a graded early warning instruction or a machining control instruction is generated and executed; wherein, the dynamic early warning threshold is determined based on the geometric complexity of the aluminum alloy shell part, the current machining stage, and the adaptive calculation of the real-time fused feature vector, and the graded early warning instruction includes at least triggering the alarm flag of the digital twin model, generating machining parameter optimization suggestions, and automatically adjusting the real-time machining parameters of the CNC machining system.

[0015] Specifically, the method of distributively embedding multiple micro piezoelectric ceramic units on the surface of the aluminum alloy housing part to form a self-sensing array conformally to the surface of the aluminum alloy housing part includes:

[0016] In the non-assembly surface area of ​​the aluminum alloy shell part, micro-dimples matching the shape of the micro piezoelectric ceramic unit are prepared by laser micromachining;

[0017] The micro piezoelectric ceramic unit is positioned in the micro recess and filled and encapsulated with high-temperature resistant insulating adhesive, so that the self-sensing array conforms to the surface of the aluminum alloy housing part and is insulated from it.

[0018] Specifically, the step of using the piezoelectric charge signal as a modulation source, modulating the modulation source through radio frequency identification backscattering to obtain a reflected radio frequency signal, and transmitting the reflected radio frequency signal as a self-sensing signal containing stress information includes:

[0019] The piezoelectric charge signal is input to a load modulation circuit and converted into a digital control signal for modulating the reflection amplitude.

[0020] The load modulation circuit changes the load impedance of a passive RFID tag antenna according to the digital control signal, modulates the amplitude of the incident RF carrier from the RF reader, generates the reflected RF signal, and transmits the reflected RF signal as a self-sensing signal containing stress information.

[0021] Specifically, receiving the self-sensing signal and simultaneously acquiring the real-time machining parameters of the CNC machining system includes:

[0022] The self-sensing signal is received by an RF reader and demodulated to obtain a stress data packet;

[0023] The real-time machining parameters, including spindle speed, feed rate, and tool position, can be read in real time through the data interface of the CNC machining system.

[0024] The stress data package and the real-time processing parameters are aligned and encapsulated based on a unified timestamp.

[0025] Specifically, the step of fusing the self-perceived signal and the real-time processing parameters through an adaptive weight allocation model to generate a real-time fused feature vector includes:

[0026] The amplitude of the self-sensing signal is normalized to obtain normalized stress characteristics;

[0027] The real-time processing parameters are numerically standardized to obtain standardized process characteristics;

[0028] The normalized stress features and the standardized process features are weighted and concatenated according to dynamic weighting coefficients to generate the real-time fused feature vector.

[0029] Specifically, the adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process, including:

[0030] Based on the process code in the real-time processing parameters, identify and output the current processing stage identifier of the CNC machining process;

[0031] According to the predefined weight mapping table, query the first weight coefficient of the self-sensing signal and the second weight coefficient of the real-time processing parameter corresponding to the current processing stage identifier.

[0032] Specifically, the step of inputting the real-time fused feature vector into a pre-trained spiking neural network model, wherein the spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value characterizing the current stress risk level of the aluminum alloy shell part, includes:

[0033] The hidden layer neurons of the spiking neural network model perform membrane potential integration based on the input real-time fused feature vector. When the membrane potential exceeds the firing threshold, a pulse event is triggered and the membrane potential is reset.

[0034] The output layer neurons of the spiking neural network model receive and accumulate the spiking events triggered by the hidden layer neurons;

[0035] Based on the accumulated values ​​of the output layer neurons, the stress risk probability value is calculated and output through a normalization function.

[0036] Specifically, the step of constructing a digital twin model of the aluminum alloy shell part and synchronizing the stress risk probability value and the real-time processing parameters to the digital twin model includes:

[0037] A three-dimensional geometric model and a physical property model of the aluminum alloy shell part are established to construct the digital twin model;

[0038] The stress risk probability value is used as a state attribute, and the real-time processing parameters are used as a process attribute. These parameters are then updated in real time to the data nodes corresponding to the digital twin model via a communication interface.

[0039] Specifically, the step of visualizing a three-dimensional thermal map on the three-dimensional model of the aluminum alloy shell part corresponding to the digital twin model based on the stress risk probability value includes:

[0040] According to the predefined color mapping rules, the stress risk probability value is mapped to the corresponding color value;

[0041] The color value is rendered to the corresponding spatial position of the three-dimensional model of the aluminum alloy shell part in the digital twin model, generating the dynamically updated three-dimensional heat map.

[0042] Specifically, the step of generating and executing graded early warning instructions or processing control instructions based on the stress risk probability value and dynamic early warning threshold includes:

[0043] The stress risk probability value is compared with a first warning threshold and a second warning threshold, wherein the second warning threshold is greater than the first warning threshold.

[0044] If the stress risk probability value exceeds the first warning threshold but does not exceed the second warning threshold, an instruction to optimize the machining parameters is generated; if the stress risk probability value exceeds the second warning threshold, an instruction to automatically adjust the real-time machining parameters of the CNC machining system is generated.

[0045] The generated instructions are sent to the corresponding human-machine interface or the CNC machining system for execution.

[0046] This application provides a method for self-sensing and intelligent early warning of stress in CNC-machined aluminum alloy shell parts. The method involves distributively embedding micro-piezoelectric ceramic units on the surface of the aluminum alloy shell part to form a conformal self-sensing array, generating piezoelectric charge signals. These signals are then transmitted after radio frequency identification backscatter modulation, carrying stress information. The method receives these signals and simultaneously acquires real-time machining parameters from the CNC machining system. An adaptive weight allocation model dynamically adjusts the fusion weight coefficients according to the current machining stage, generating a real-time fusion feature vector. This vector is input into a pre-trained pulse neural network model, which outputs a stress risk probability value. This value is then synchronized to a digital twin model for 3D heatmap visualization. Based on the stress risk probability value and a dynamically calculated early warning threshold adapted to the part's geometric complexity, the method generates and executes graded early warning or machining control commands. This achieves real-time, passive, distributed sensing and dynamic risk assessment of machining stress status, improving the timeliness of early warnings and the autonomous controllability of machining. Attached Figure Description

[0047] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a flowchart illustrating the self-sensing and intelligent early warning method for stress in CNC-machined aluminum alloy shell parts provided in this application.

[0049] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0052] In this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0053] The method for self-sensing and intelligent early warning of stress in CNC-machined aluminum alloy shell parts provided in this application generates piezoelectric charge signals by forming a conformal array of micro piezoelectric ceramic units distributedly embedded on the surface of the part; transmitting stress information signals via radio frequency identification backscatter modulation; receiving signals and synchronously acquiring real-time machining parameters, and using an adaptive weight allocation model to generate a real-time fused feature vector; inputting a pulse neural network model to output a stress risk probability value; synchronizing to a digital twin model for visualization; and generating and executing graded early warning or machining control commands based on the stress risk probability value and dynamic early warning threshold.

[0054] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0055] Figure 1The flowchart illustrates the self-sensing and intelligent early warning method for stress in CNC-machined aluminum alloy shell parts provided in this application. Figure 1 As shown, this embodiment provides a method for self-sensing and intelligent early warning of stress in CNC-machined aluminum alloy shell parts. The method includes:

[0056] S101: Multiple micro piezoelectric ceramic units are distributedly embedded on the surface of the aluminum alloy housing part to form a self-sensing array conforming to the surface of the aluminum alloy housing part.

[0057] The micro piezoelectric ceramic unit is used to generate piezoelectric charge signals based on the stress applied during CNC machining.

[0058] Specifically, the method of distributively embedding multiple micro piezoelectric ceramic units on the surface of the aluminum alloy shell part to form a self-sensing array conforming to the surface of the aluminum alloy shell part includes: preparing micro-recesses matching the shape of the micro piezoelectric ceramic units in the non-assembly surface area of ​​the aluminum alloy shell part by laser micromachining; positioning the micro piezoelectric ceramic units in the micro-recesses and filling and encapsulating them with high-temperature resistant insulating adhesive, so that the self-sensing array conforms to the surface of the aluminum alloy shell part and is insulated from it.

[0059] The specific implementation of step S101 includes the following sub-steps:

[0060] S1011: Determine the embedding location of the micro piezoelectric ceramic unit.

[0061] Based on the 3D CAD model of the aluminum alloy shell part, all non-assembly surface areas on the part are identified. Non-assembly surface areas refer to surfaces that, in the final assembled state, do not form a tight fit or load-bearing interface with other components through bolting, gluing, welding, or other methods. On the selected non-assembly surface areas, according to a preset distribution density (e.g., 1 to 4 units per square centimeter), avoiding the toolpaths of subsequent CNC machining, the precise center coordinates of each micro piezoelectric ceramic unit are determined. The micro piezoelectric ceramic unit is preferably a rectangular or circular sheet made of lead zirconate titanate (PZT-5H) material, with dimensions ranging from 0.5 mm to 2 mm in length, 0.5 mm to 2 mm in width, and 0.1 mm to 0.3 mm in thickness.

[0062] S1012: Micro-dimples are fabricated using laser micromachining.

[0063] Based on the center coordinates determined in S1011, a picosecond ultraviolet laser is used to perform laser micromachining on the non-assembly surface area of ​​the aluminum alloy housing part. The laser processing parameters are set as follows: wavelength 355 nm, pulse width 10 picoseconds, repetition frequency 100 kHz, and single pulse energy 30 microjoules. The laser beam is controlled to scan along a preset contour path to remove material by ablation, forming micro-recesses that match the shape of the micro-piezoelectric ceramic unit. The depth of the micro-recesses is 10 to 30 micrometers greater than the thickness of the micro-piezoelectric ceramic unit to ensure packaging space. After processing, the micro-recesses are ultrasonically cleaned with anhydrous ethanol to remove slag and dust.

[0064] S1013: Positioning and packaging of miniature piezoelectric ceramic units.

[0065] Using a vacuum pen or precision robotic arm, each micro piezoelectric ceramic unit is picked up and precisely placed into its corresponding micro-recess. The polarization direction of the micro piezoelectric ceramic unit is ensured to be perpendicular to the part surface and aligned. Then, a high-temperature resistant insulating adhesive is used to fill and encapsulate the micro-recesses. This high-temperature resistant insulating adhesive is a silicone potting compound that can withstand continuous operating temperatures above 200 degrees Celsius after curing. The high-temperature resistant insulating adhesive is injected into the micro-recesses using a dispensing device until the adhesive completely covers the micro piezoelectric ceramic unit and slightly overflows the edges of the recess. Subsequently, the part is placed in an 80-degree Celsius oven for 2 hours to cure the insulating adhesive completely. After curing, the encapsulated area is lightly polished using a precision grinding device to make the surface of the insulating adhesive flush with the original surface of the part, thereby ensuring that the self-sensing array is conformally aligned with and insulated from the surface of the aluminum alloy housing part.

[0066] This step involves creating micro-recesses in non-critical areas (non-assembly surfaces) of the part using laser micromachining technology, precisely embedding micro-piezoelectric ceramic units within them, and finally encapsulating them using high-temperature resistant insulating adhesive. This process creates a self-sensing array that perfectly conforms to the part's surface. Its advantages are: First, the sensor unit is integrated with the part body, without altering the part's macroscopic structure or occupying additional space, achieving true "embedded" sensing and avoiding interference from external sensors on the processing space and the dynamic characteristics of the part. Second, the insulating encapsulation ensures the electrical safety and long-term reliability of the sensor unit in harsh environments such as processing coolant and metal shavings. Third, the conformal design avoids stress concentration or flow field disturbances caused by sensor protrusions or recesses, laying the physical foundation for accurate sensing of stress changes caused by the processing process itself, which is a prerequisite for achieving "self-sensing."

[0067] S102: The piezoelectric charge signal is used as a modulation source, and the modulation source is modulated by radio frequency identification backscattering to obtain a reflected radio frequency signal, and the reflected radio frequency signal is sent as a self-sensing signal containing stress information.

[0068] Specifically, the step of using the piezoelectric charge signal as a modulation source, modulating the modulation source through radio frequency identification backscattering to obtain a reflected radio frequency signal, and transmitting the reflected radio frequency signal as a self-sensing signal containing stress information includes: inputting the piezoelectric charge signal to a load modulation circuit to convert it into a digital control signal for modulating the reflection amplitude; the load modulation circuit changing the load impedance of a passive radio frequency identification tag antenna according to the digital control signal to modulate the amplitude of the incident radio frequency carrier from the radio frequency reader, generating the reflected radio frequency signal, and transmitting the reflected radio frequency signal as a self-sensing signal containing stress information.

[0069] The specific implementation of step S102 includes the following sub-steps:

[0070] S1021: Conditioning and digitizing the piezoelectric charge signal.

[0071] The piezoelectric charge signal Qi generated by the i-th micro piezoelectric ceramic unit in S101 is first input to a signal conditioning circuit integrated on a passive RFID tag. This signal conditioning circuit is implemented by a charge amplifier circuit composed of an operational amplifier, whose function is to convert the charge signal Qi into a voltage signal Vi. The conversion relationship is determined by the feedback capacitor Cf, i.e., Vi = -Qi / Cf. The voltage signal Vi is then input to a voltage comparator. The voltage comparator compares the voltage signal Vi with a preset reference voltage Vref. When the absolute value of the voltage signal Vi exceeds the reference voltage Vref, the voltage comparator outputs a high-level digital control signal Di(t); otherwise, it outputs a low-level signal. The digital control signal Di(t) is a binary sequence whose pulse density or pulse width is positively correlated with the amplitude of the piezoelectric charge signal Qi, thereby encoding stress information in the digital control signal Di(t).

[0072] S1022: Change the antenna impedance through the load modulation circuit.

[0073] The digital control signal Di(t) is fed into the drive port of the load modulation circuit. The core of the load modulation circuit is a CMOS analog switch controlled by the digital control signal Di(t), which is connected in parallel in the resonant circuit of the passive RFID tag antenna coil. When the digital control signal Di(t) is high, the CMOS analog switch is turned on, connecting an additional resistor Rmod in parallel to the antenna resonant circuit, significantly reducing the quality factor of the antenna circuit and thus increasing the equivalent load impedance of the antenna. When the digital control signal Di(t) is low, the CMOS analog switch is turned off, and the antenna resonant circuit returns to a high-impedance state.

[0074] S1023: Generates a reflection modulation signal using the incident radio frequency carrier.

[0075] A radio frequency (RF) reader deployed near the working area of ​​a CNC machine tool continuously transmits an incident RF carrier at a frequency of fc (e.g., 915 MHz). This incident RF carrier is received by the antenna of a passive RFID tag, providing the energy required for the tag's chip to operate. The tag's reflection characteristics of the incident RF carrier change according to the change in the antenna load impedance in S1022. Specifically, when the CMOS analog switch is turned on, the antenna impedance mismatch increases, enhancing the reflection amplitude of the incident RF carrier; when the switch is turned off, the reflection amplitude decreases. This process achieves amplitude shift keying (APS) modulation of the reflected RF signal amplitude using a digital control signal Di(t). The envelope shape of the modulated reflected RF signal Si(t) is consistent with the waveform of the digital control signal Di(t), thus carrying stress information in the amplitude variation of the reflected RF signal Si(t).

[0076] S1024: Send a self-sensing signal containing stress information.

[0077] The reflected radio frequency signal Si(t), modulated by the load, is radiated into space through the antenna of the passive RFID tag, forming the self-sensing signal containing stress information. Since the tag itself has no battery, its energy comes entirely from the incident radio frequency carrier of the RFID reader; therefore, the entire signal generation and transmission process is passive. The signals from multiple distributed tags are distinguished in the time or frequency domain using multiple access technology and are uniformly received by the RFID reader.

[0078] This step creatively integrates piezoelectric sensing with passive RFID backscatter communication technology, achieving a completely passive process from stress information generation to wireless transmission. Its core functions are: First, by amplifying the charge and comparing the voltage, the weak analog piezoelectric charge signal is reliably converted into a digital control signal suitable for wireless transmission, solving the problem of direct processing and transmission of low-energy signals; Second, utilizing mature load modulation technology, the incident carrier is amplitude modulated by changing the antenna impedance—a low-power method—to efficiently encode the digital control signal (i.e., stress information) into the reflected radio frequency signal. This process consumes extremely low power and can be achieved entirely by the reader; Third, the final generated "self-sensing signal" is a standard modulated radio frequency signal, compatible with existing RFID communication protocols and hardware, providing a stable, reliable, and standard-equippable wireless data source for subsequent steps. This solution fundamentally eliminates the wiring or batteries required by traditional active sensors, achieving integrated, passive integration of the sensing unit and the part body, which is a key technological step in achieving the invention's objective of "self-sensing without interfering with processing."

[0079] S103: Receive the self-sensing signal and synchronously acquire the real-time machining parameters of the CNC machining system.

[0080] The CNC machining system is used to indicate and control the CNC machining process.

[0081] Specifically, receiving the self-sensing signal and synchronously acquiring the real-time machining parameters of the CNC machining system includes: receiving the self-sensing signal through an RF reader and demodulating it to obtain a stress data packet; reading the real-time machining parameters, including spindle speed, feed rate, and tool position, in real time through the data interface of the CNC machining system; and aligning and encapsulating the stress data packet and the real-time machining parameters based on a unified timestamp.

[0082] The specific implementation of step S103 includes the following sub-steps:

[0083] S1031: Receive and demodulate the self-sensing signal to obtain the stress data packet.

[0084] An RFID reader deployed near the processing area continuously listens for a specific frequency band (e.g., the UHF band 920-925 MHz). The RFID reader receives reflected RFID signals Si(t) backscattered from one or more passive RFID tags in S102. The RFID reader's internal RFID receiving circuit down-converts and filters the reflected RFID signal Si(t) to obtain an intermediate frequency (IF) signal. Then, the RFID reader's digital signal processing unit performs coherent demodulation or envelope detection on the IF signal to recover the waveform of the digital control signal Di(t) carried on the amplitude of the reflected RFID signal Si(t). Next, the RFID reader or its connected central processing unit decodes the recovered digital signal waveform. The decoding process translates the pulse sequence into a binary data stream according to preset encoding rules (e.g., Manchester encoding or PIE encoding), and parses the binary data stream into a sequence of raw stress values ​​corresponding to the spatial location (i.e., the corresponding micro-piezoelectric ceramic unit) based on the unique identification code pre-assigned to each passive RFID tag. Finally, the raw stress values ​​parsed from all labels at each sampling time are packaged together with the spatial location code and timestamp of the corresponding label to form a structured stress data packet. The data structure of this stress data packet can be defined as a list of {timestamp, [sensor_id, stress_value]}.

[0085] S1032: Reads real-time machining parameters from the CNC machining system via a data interface.

[0086] The central processing unit (CPU) establishes a connection with the CNC machining system via industrial Ethernet or fieldbus using standard industrial data communication protocols (such as OPCUA, MTConnect, or proprietary APIs provided by the CNC system manufacturer). The CPU sends data request commands to the CNC machining system through this connection at a preset sampling frequency (e.g., 100 Hz). The CNC machining system responds to the request and returns a response data stream containing the key machining status parameters at the current moment. The CPU parses and extracts three key types of real-time machining parameters from the response data stream: spindle speed (rpm), feed rate (mm / min), and the current position of the tool in the workpiece coordinate system (X, Y, Z coordinates, mm). Simultaneously, the CPU timestamps the acquisition time for these real-time machining parameters.

[0087] S1033: Based on unified timestamp alignment and encapsulation of stress data packets and real-time processing parameters.

[0088] The central processing unit (CPU) maintains a high-precision system clock as a unified time reference. The CPU compares the timestamps in the stress data packets generated in S1031 with the timestamps of the real-time processing parameters acquired in S1032. For each stress data packet, the CPU searches for the real-time processing parameter dataset with the smallest absolute time difference in the time series of the real-time processing parameters. A linear interpolation algorithm is used to perform time alignment compensation on the found real-time processing parameters. Specifically, let the timestamp of the stress data packet be t_stress, and the timestamps of two adjacent real-time processing parameter sampling points be t_param1 and t_param2 (t_param1 < t_stress < t_param2), with corresponding parameter values ​​P1 and P2 respectively. Then, the aligned real-time processing parameter value P_aligned is calculated using the linear interpolation formula: P_aligned = P1 + ((t_stress - t_param1) / (t_param2 - t_param1)) * (P2 - P1). Finally, the original stress value sequence in the time-aligned stress data packet is combined with the aligned real-time machining parameters (spindle speed, feed rate, tool position) and encapsulated into a new synchronization data frame. The data structure of this synchronization data frame can be defined as: {sync_timestamp, [sensor_id, stress_value], spindle_speed, feed_rate, tool_position_x, tool_position_y, tool_position_z}. This synchronization frame encapsulates time-aligned multi-source data. The sync_timestamp field represents the precise timestamp after data alignment; [sensor_id, stress_value] is a list where sensor_id represents the unique identifier of a passive RFID tag emitting a self-sensing signal, corresponding to the spatial position of a micro piezoelectric ceramic unit; stress_value represents the original stress value sensed and demodulated by this unit; spindle_speed and feed_rate represent the spindle speed and feed rate, respectively; and tool_position_x, tool_position_y, and tool_position_z together represent the three-dimensional coordinate position of the tool in the workpiece coordinate system.

[0089] This step serves as a crucial bridge connecting the physical and digital information worlds, its core function being the precise synchronization and fusion preprocessing of heterogeneous data. First, a radio frequency reader performs professional radio frequency reception, demodulation, and decoding of the backscattered signal, reliably restoring the airborne wireless signal into stress data packets that can be processed by a computer, completing the first conversion from physical signals to digital information. Second, the internal state parameters of the CNC system are acquired in real time through a standard industrial interface, breaking down information barriers and obtaining the precise context of the machining process. Finally, using a high-precision unified clock-based timestamp alignment and linear interpolation algorithm, asynchronously arriving stress sensing data from different sources are precisely matched with the machining process data in the time dimension, generating synchronized data frames with strict temporal correspondence. This step eliminates analytical errors caused by asynchronous data acquisition, providing high-quality, spatiotemporally aligned input for accurate multi-source data fusion in the subsequent adaptive weight allocation model in S104, a fundamental prerequisite for ensuring the accuracy and timeliness of the entire intelligent early warning system's analysis results.

[0090] S104: The self-sensing signal and the real-time machining parameters are fused using an adaptive weight allocation model to generate a real-time fused feature vector. The adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process.

[0091] Specifically, the step of fusing the self-sensing signal and the real-time processing parameters through an adaptive weight allocation model to generate a real-time fused feature vector includes: normalizing the amplitude of the self-sensing signal to obtain normalized stress features; standardizing the real-time processing parameters to obtain standardized process features; and weighting and concatenating the normalized stress features and the standardized process features according to dynamic weight coefficients to generate the real-time fused feature vector.

[0092] Specifically, the adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process, including: identifying and outputting the current machining stage identifier of the CNC machining process based on the process code in the real-time machining parameters; and querying the first weight coefficient of the self-sensing signal and the second weight coefficient of the real-time machining parameters corresponding to the current machining stage identifier according to a predefined weight mapping table.

[0093] The specific steps in step S104 during implementation include:

[0094] The specific implementation of S104 includes the following steps:

[0095] S1041: The amplitude of the self-sensing signal is normalized to obtain the normalized stress characteristics.

[0096] From the synchronous data frame output by S103, the raw stress values ​​(stress_value) of all sensors (micro piezoelectric ceramic units) at the same sampling time (sync_timestamp) are extracted to form a raw stress vector S_raw = [s1,s2, ..., sn], where n is the number of sensors and si represents the raw stress value of the i-th sensor. To eliminate the influence of sensitivity differences between different sensors and signal baseline drift, the max-min normalization method is used to process the raw stress vector S_raw. First, based on the historical data of the sensor during a static period before processing begins, the minimum stress value (s_min) and maximum stress value (s_max) are calculated. Then, the following formula is applied to normalize each raw stress value (si) at the current time:

[0097] si_norm = (si - s_min) / (s_max - s_min).

[0098] This calculation maps the stress value of each sensor to the interval [0, 1] (truncated to 0 or 1 if the signal exceeds the historical range), resulting in a normalized stress vector S_norm = [s1_norm, s2_norm, ..., sn_norm]. Finally, the L2 norm (Euclidean norm) of the S_norm vector is calculated as the comprehensive normalized stress characteristic F_stress at that moment, i.e., F_stress = ||S_norm||_2 = sqrt(s1_norm^2 + s2_norm^2 + ... + sn_norm^2). F_stress is a scalar representing the overall relative stress level of the part at that moment.

[0099] S1042: Perform numerical standardization on real-time processing parameters to obtain standardized process characteristics.

[0100] Real-time machining parameters are extracted from the same synchronized data frame: spindle speed N (unit: rpm), feed rate F (unit: mm / min), and tool position (X, Y, Z) (unit: mm). Since these parameters have different dimensions and numerical ranges, Z-score normalization is used to transform them to a similar scale. First, based on historical machining data or process specifications, the historical mean (μ) and standard deviation (σ) of each parameter are calculated. For example, for spindle speed N, there are a mean μ_N and a standard deviation σ_N. Then, the normalization calculation is performed on each parameter at the current moment:

[0101] N_std = (N - μ_N) / σ_N;

[0102] F_std = (F - μ_F) / σ_F;

[0103] X_std = (X - μ_X) / σ_X;

[0104] Y_std = (Y - μ_Y) / σ_Y;

[0105] Z_std = (Z - μ_Z) / σ_Z.

[0106] The five standardized parameters are concatenated into a standardized process feature vector F_process = [N_std,F_std, X_std, Y_std, Z_std]. In the formula, N, F, X, Y, and Z represent the spindle speed, feed rate, and tool position coordinates in the X, Y, and Z axes at the current moment, extracted from the synchronous data frame, respectively; μ_N, μ_F, μ_X, μ_Y, and μ_Z are the historical averages of the corresponding parameters calculated based on historical machining data or process specifications; σ_N, σ_F, σ_X, σ_Y, and σ_Z are the historical standard deviations of the corresponding parameters; and N_std, F_std, X_std, Y_std, and Z_std obtained after Z-score standardization are the standardized values ​​after eliminating the influence of dimensions.

[0107] S1043: Identify the current processing stage and query the dynamic weight coefficients.

[0108] The process codes (G-codes / M-codes) in the real-time machining parameters are parsed from the synchronous data frames. A predefined rule parser identifies the currently executing machining operation type based on the process code and maps it to a preset machining stage identifier (Phase_ID). For example, codes G01 / G02 / G03 might correspond to the "contour milling stage," and G81 / G83 might correspond to the "drilling stage." Subsequently, a predefined weight mapping table is queried. This weight mapping table is stored in key-value pairs, where the key is the machining stage identifier (Phase_ID), and the value is a tuple (w_stress, w_process) containing two weight coefficients. w_stress is the fusion weight of the normalized stress feature F_stress (first weight coefficient), and w_process is the fusion weight of the normalized process feature vector F_process (second weight coefficient). The two weight coefficients satisfy the relationship w_stress + w_process = 1. For example, in the "rough machining stage", (w_stress=0.3, w_process=0.7) might be set to emphasize the dominance of process parameters; in the "finish machining stage", (w_stress=0.7, w_process=0.3) might be set to emphasize the sensitivity of stress perception.

[0109] S1044: Weighted concatenation is performed according to dynamic weight coefficients to generate a real-time fused feature vector.

[0110] The scalar F_stress and the vector F_process are weighted and concatenated to generate the real-time fused feature vector F_fused. First, the scalar F_stress is multiplied by the corresponding weight coefficient w_stress to obtain the weighted stress feature F_stress_weighted = w_stress * F_stress. Then, each element in the vector F_process is multiplied by the corresponding weight coefficient w_process to obtain the weighted process feature vector F_process_weighted = w_process * F_process. Finally, the weighted scalar F_stress_weighted and the weighted vector F_process_weighted are concatenated into a new vector, i.e., F_fused = concat(F_stress_weighted, F_process_weighted) = [w_stress * F_stress, w_process * N_std, w_process * F_std, w_process * X_std, w_process * Y_std, w_process * Z_std]. The F_fused vector is the result of adaptive fusion and will be used as the input to the spiking neural network model in S105.

[0111] This step is the core of multi-source information fusion and feature engineering. Its function is to transform two types of data—physical signals (stress) and process parameters (rotation speed, feed, position)—which have different properties and dimensions, into a standardized feature vector that reflects the overall state of the processing and is suitable for neural network input. Its effects are reflected in three aspects: First, through max-min normalization and Z-score standardization, the dimensional differences and distribution biases of the original data are eliminated, improving the stability and convergence speed of subsequent model training. Second, through a dynamic weight allocation mechanism based on process codes, the fusion strategy is adaptively adjusted, allowing the model to flexibly emphasize stress or process information at different processing stages. This is more accurate in characterizing the differences in stress generation mechanisms at different process stages than a fixed-weight fusion method. Finally, the generated real-time fused feature vector F_fused has a regular structure and fixed dimensions, providing high-quality input for subsequent temporal event processing of the spiking neural network. It is a crucial bridge connecting data acquisition and intelligent evaluation, directly determining the accuracy and reliability of the entire system's risk assessment.

[0112] S105: Input the real-time fused feature vector into a pre-trained spiking neural network model. The spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value that characterizes the current stress risk level of the aluminum alloy shell part.

[0113] Specifically, the step of inputting the real-time fused feature vector into a pre-trained spiking neural network model, wherein the spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value characterizing the current stress risk level of the aluminum alloy shell part, includes: the hidden layer neurons of the spiking neural network model performing membrane potential integration based on the input real-time fused feature vector; triggering a pulse event and resetting the membrane potential when the membrane potential exceeds a firing threshold; the output layer neurons of the spiking neural network model receiving and accumulating the pulse events triggered by the hidden layer neurons; and calculating and outputting the stress risk probability value based on the accumulated value of the output layer neurons through a normalization function.

[0114] The specific implementation of S105 includes the following steps:

[0115] S1051: Construct and load the pre-trained spiking neural network model.

[0116] The described spiking neural network model employs a three-layer feedforward structure, comprising an input layer, a hidden layer, and an output layer. The input layer contains six neurons, corresponding to the six dimensions of the real-time fused feature vector F_fused (one weighted stress feature + five weighted process features). The hidden layer contains 32 neurons using a Leaky Integrate-and-Fire (LIF) model. The output layer contains one neuron using the same LIF model. Full connections exist between the input and hidden layers, and between the hidden and output layers, with each connection having a trainable synaptic weight. The model has been pre-trained offline using historical processing data (containing a large number of real-time fused feature vector F_fused sequences under normal and abnormal stress states and their corresponding stress risk labels). Training employs a Backpropagation Through Time (BPTT) algorithm combined with an alternative gradient method, aiming to minimize the cross-entropy loss between the model's predicted risk probability and the actual risk label, thereby optimizing all synaptic weights and neuron parameters.

[0117] S1052: Hidden layer neurons perform membrane potential integration and impulse event triggering.

[0118] At each discrete time step t (e.g., a time step of 1 millisecond), the real-time fused feature vector F_fused(t) at the current moment is input into the spiking neural network model. For the j-th LIF neuron in the hidden layer, its membrane potential V_j(t) is updated according to the following formula:

[0119] V_j(t) = τ * V_j(t-1) + Σ (w_ij * X_i(t)).

[0120] Where τ is the membrane potential decay time constant (e.g., τ = 0.9), V_j(t-1) is the membrane potential of the neuron in the previous time step, w_ij is the synaptic weight from the i-th input layer neuron to the hidden layer neuron, and Xi(t) is the input value of the i-th input layer neuron at the current time (i.e., the i-th component of F_fused(t)). If the updated membrane potential V_j(t) exceeds the firing threshold V_th of the neuron (e.g., V_th = 1.0), the hidden layer neuron triggers a pulse event at time t (output is 1), and its membrane potential V_j(t) is then reset to the resting potential V_reset (e.g., V_reset = 0); if the threshold is not exceeded, there is no pulse output (output is 0). This process embodies a sparse pulse event-driven mechanism, that is, only a few neurons with membrane potentials exceeding the threshold will generate computational overhead (firing pulses) at specific times.

[0121] S1053: Output layer neurons receive and accumulate pulses.

[0122] The LIF neurons in the output layer receive pulse inputs from all 32 hidden layer neurons. At time step t, the update of the membrane potential V_out(t) of the output layer neurons is similar to that of the hidden layer neurons in S1052, but its input X_j(t) is the output (0 or 1) of hidden layer neuron j at time t. The output layer neurons also decide whether to fire a pulse based on whether the membrane potential exceeds the firing threshold V_th_out. Within a fixed time window T (e.g., T = 100 time steps, corresponding to 100 milliseconds of physical time), the total number of pulses fired by the output layer neurons is accumulated and denoted as the pulse accumulation value S.

[0123] S1054: Calculate and output the stress risk probability value.

[0124] The accumulated pulse value S obtained within the time window T is input into a normalization function, which maps it to a scalar value in the interval [0, 1], namely the stress risk probability value P_risk. The normalization function uses the Sigmoid function, and its calculation formula is as follows:

[0125] P_risk = 1 / (1 + exp(-k * (S - S0))).

[0126] Where S is the accumulated pulse value, S0 is the center offset parameter of the sigmoid function, and k is the steepness coefficient of the function. Parameters S0 and k are determined together during the model pre-training phase. The function output P_risk is the probability value characterizing the stress risk level of the current aluminum alloy shell part. The closer its value is to 1, the higher the risk of stress exceeding limits or processing abnormalities; the closer it is to 0, the safer the processing condition.

[0127] This step utilizes a spiking neural network to simulate the information processing mechanism of biological neurons, achieving a core component for efficient and energy-saving risk assessment. Its function is to transform the real-time fused feature vector F_fused, reflecting the overall processing status, into a continuous scalar output with a clear physical meaning (risk probability) through the temporal integration of neuronal membrane potentials and the discrete pulse firing mechanism in the spiking neural network. Its significant effects are reflected in three aspects: First, the "event-driven" nature of the spiking neural network means that pulses are only triggered for calculation when the input features undergo significant changes (corresponding to membrane potentials exceeding a threshold). Compared to traditional artificial neural networks that perform dense matrix operations every moment, it has a higher energy efficiency ratio when processing continuous temporal fused feature vectors and is more suitable for edge computing deployment. Second, by learning from historical data, the model can capture the complex nonlinear spatiotemporal patterns implicit in the fused feature vector that are highly correlated with stress risk, such as the subtle abnormal fluctuations in stress characteristics under specific process parameter combinations, thereby achieving accurate and forward-looking assessment of risk probability. Finally, the output stress risk probability value P_risk is a continuous value between 0 and 1, which provides richer and more refined decision-making basis for subsequent steps (such as visualization in S106 and hierarchical early warning in S107) than simple binary judgment, enabling the early warning system to distinguish different levels of risk and achieve more flexible and intelligent control.

[0128] S106: Construct a digital twin model of the aluminum alloy shell part and synchronize the stress risk probability value and the real-time processing parameters to the digital twin model, and perform a three-dimensional thermal visualization on the three-dimensional model of the aluminum alloy shell part corresponding to the digital twin model according to the stress risk probability value.

[0129] Specifically, the step of constructing a digital twin model of the aluminum alloy shell part and synchronizing the stress risk probability value and the real-time processing parameters to the digital twin model includes: establishing a three-dimensional geometric model and a physical property model of the aluminum alloy shell part to construct the digital twin model; using the stress risk probability value as a state attribute and the real-time processing parameters as a process attribute, and updating them in real time to the data nodes corresponding to the digital twin model through a communication interface.

[0130] Specifically, the step of visualizing a three-dimensional heat map on the three-dimensional model of the aluminum alloy shell part corresponding to the stress risk probability value in the digital twin model includes: mapping the stress risk probability value to a corresponding color value according to a predefined color mapping rule; rendering the color value to the corresponding spatial position of the three-dimensional model of the aluminum alloy shell part in the digital twin model, and generating the dynamically updated three-dimensional heat map.

[0131] The specific implementation of S106 includes the following steps:

[0132] S1061: Construct a digital twin model of an aluminum alloy shell part.

[0133] In the digital twin development platform, the 3D CAD model file (such as STEP or IGES format) of the aluminum alloy shell part is imported as the geometric model basis of the digital twin. Subsequently, physical properties are added to the 3D geometric model, specifically including: defining material properties for the model (such as aluminum alloy grade, density, and elastic modulus), and creating and associating a virtual "sensor data node" at the corresponding surface location of the 3D model for the embedding position of each micro piezoelectric ceramic unit determined in S101. Each sensor data node is a data structure used to store and update the stress risk probability value P_risk corresponding to that location. Simultaneously, text or dashboard controls are created in the digital twin scene to display real-time machining parameters (spindle speed, feed rate, tool position). This virtual entity, integrating geometric and physical properties and data interfaces, constitutes the digital twin model described in this step.

[0134] S1062: Synchronize the stress risk probability value with the real-time processing parameters to the digital twin model.

[0135] A real-time data communication link is established between the central processing unit (the calculation unit executing S104 and S105) and the digital twin development platform. This link can use the WebSocket protocol or the OPC UA protocol. In each calculation cycle (e.g., every 100 milliseconds, consistent with the output frequency of S105), the central processing unit sends the stress risk probability value P_risk calculated at the current moment to the corresponding sensor data node in the digital twin development platform via the communication link. Simultaneously, the real-time machining parameters (spindle speed N, feed rate F, tool position X, Y, Z) acquired from S103 at the same time are sent to the corresponding text or dashboard controls in the digital twin scene. The data receiving module of the digital twin development platform monitors this communication link, parses the received data, and immediately updates the stress risk probability value P_risk stored in the sensor data node and the real-time machining parameter values ​​in the display controls, achieving real-time synchronization of the physical world state with the virtual model.

[0136] S1063: Generate a three-dimensional heat map based on the stress risk probability value.

[0137] In the digital twin development platform, a vertex-shading-based material is configured for the 3D model of the aluminum alloy shell part. A shader script is written that performs the following operations: for each vertex on the surface of the 3D model, based on the stress risk probability value P_risk stored in its associated sensor data node, the corresponding color value (RGB value) is queried or calculated according to a predefined color mapping rule. The color mapping rule is a mapping function from the [0, 1] interval to a color spectrum. For example, linear interpolation mapping is used: when P_risk = 0, it is mapped to blue (RGB: 0, 0, 255); when P_risk = 0.5, it is mapped to yellow (RGB: 255, 255, 0); when P_risk = 1, it is mapped to red (RGB: 255, 0, 0). For a P_risk value between 0 and 1, its color is calculated using a linear interpolation formula: R = 255 * min(2 * P_risk, 1), G = 255 * min(2 * (0.5 - |P_risk - 0.5|), 1), B = 255 * min(2 * (1 - P_risk), 1). The shader script assigns the calculated color value to the corresponding vertex.

[0138] S1064: Render and display a dynamically updated 3D heatmap.

[0139] The graphics rendering engine of the digital twin development platform calls the aforementioned shader script to render the 3D model of the aluminum alloy shell part in real time. Since the stress risk probability value P_risk in the sensor data nodes is dynamically updated, the color of the rendered model surface also changes dynamically, forming a 3D heatmap that intuitively reflects the distribution and evolution of stress risk at various points on the part. This 3D heatmap, along with controls displaying real-time processing parameters, is displayed on the digital twin monitoring interface. Operators can rotate and zoom the 3D model to observe the heatmap from any angle, thereby intuitively grasping the overall stress risk status and spatial distribution of the part during processing.

[0140] This step, by constructing and driving a digital twin model, transforms the abstract data (stress risk probability values, processing parameters) calculated in the previous steps into a highly visualized 3D dynamic image, achieving a virtual-real fusion and intuitive perception of the processing status. Its core function is to provide an intuitive and immersive monitoring window for human-computer interaction. The effects are manifested in: First, the 3D heatmap directly maps the spatially distributed risk probabilities onto the 3D model of the part in the form of color gradients, making high-risk areas of stress concentration immediately apparent, far exceeding the information transmission efficiency of traditional 2D charts or numerical lists; Second, the synchronous presentation of real-time processing parameters and the risk heatmap establishes an intuitive correlation between process operation and stress state, helping operators or the system quickly locate problematic process steps; Third, the dynamically updated digital twin constitutes a continuously evolving "digital archive" of the processing process, providing a visual carrier for subsequent process backtracking, optimization, and knowledge accumulation, greatly improving the status perception capability and decision support level of the entire early warning system.

[0141] S107: Generate and execute graded early warning instructions or processing control instructions based on the stress risk probability value and dynamic early warning threshold.

[0142] The dynamic early warning threshold is determined based on the geometric complexity of the aluminum alloy shell part, the current processing stage, and the adaptive calculation of the real-time fused feature vector. The hierarchical early warning instruction includes at least triggering the alarm identifier of the digital twin model, generating processing parameter optimization suggestions, and automatically adjusting the real-time processing parameters of the CNC machining system.

[0143] Specifically, the step of generating and executing graded early warning instructions or machining control instructions based on the stress risk probability value and dynamic early warning threshold includes: comparing the stress risk probability value with a first early warning threshold and a second early warning threshold, wherein the second early warning threshold is greater than the first early warning threshold; if the stress risk probability value exceeds the first early warning threshold but does not exceed the second early warning threshold, generating an instruction for optimizing the machining parameters; if the stress risk probability value exceeds the second early warning threshold, generating an instruction for automatically adjusting the real-time machining parameters of the CNC machining system; and sending the generated instruction to the corresponding human-machine interface or the CNC machining system for execution.

[0144] This step, based on the stress risk probability value P_risk calculated by S105, uses an adaptive dynamic decision-making mechanism to determine the risk level of the current processing state and trigger corresponding early warning or control actions, forming a closed loop of perception-analysis-decision-execution. The implementation of this step is based on the established early warning rules and processing parameter adjustment strategy library, and the establishment of a reliable command communication link between the central processing unit, the digital twin monitoring interface, and the CNC machining system.

[0145] The specific implementation of S107 includes the following steps:

[0146] S1071: Adaptive calculation of dynamic early warning threshold.

[0147] The geometric complexity factor G is calculated based on the geometric characteristics of the aluminum alloy shell part. The geometric complexity factor G is obtained by multiplying the ratio of the part's surface area to its volume by a number of features reflecting its structural complexity (such as the number of holes, cavities, and thin walls). The calculation formula is: G = (part surface area / part volume) * number of structural features. The current processing stage identifier Phase_ID is obtained from S104 and converted into a stage risk baseline coefficient K_phase according to a predefined mapping table. This coefficient reflects the inherent sensitivity of different processing stages (such as roughing and finishing) to stress risk; for example, the K_phase value for the finishing stage is greater than that for the roughing stage. The standard deviation of all components is extracted from the real-time fused feature vector F_fused generated in S104, denoted as the feature fluctuation σ, to characterize the stability of the current operating condition. Finally, the first warning threshold Th1 and the second warning threshold Th2 are dynamically calculated using the following formula:

[0148] Th1 = α * G * K_phase + β * σ;

[0149] Th2 = γ * Th1.

[0150] Here, α and β are weighting coefficients used to balance the contribution of geometric complexity and feature fluctuations to the threshold, and γ is the proportional coefficient of the second threshold relative to the first threshold (γ > 1, for example, γ = 1.5). The mapping relationship between coefficients α, β, γ and K_phase is determined by collecting a large amount of historical processing data and corresponding stress risk results, and adjusting the parameters using linear regression or optimization algorithms (such as gradient descent) to achieve the best judgment effect on historical risk data, thereby finally determining the coefficients.

[0151] S1072: Hierarchical comparison and instruction decision-making.

[0152] The stress risk probability value P_risk output by S105 at the current moment is compared in real time with the first warning threshold Th1 and the second warning threshold Th2 calculated by S1071.

[0153] If P_risk > Th1 and P_risk ≤ Th2, it is determined to be a Level 1 warning (yellow warning). The system generates a machining parameter optimization suggestion instruction. The generation logic of this instruction is as follows: based on the current real-time machining parameters (spindle speed N, feed rate F) and the current machining stage identifier Phase_ID, the pre-stored "machining parameter optimization strategy library" is queried. This strategy library stores parameter adjustment experience that has been verified to effectively reduce stress risk under different stages and different parameter combinations, such as "when Phase_ID is finish milling, if P_risk is at Level 1 warning, it is recommended to reduce the feed rate F by 10%".

[0154] If P_risk > Th2, it is classified as a Level 2 warning (red warning). The system generates an automatic adjustment instruction for machining parameters. This instruction will directly include a set of specific, conservative safety process parameters, such as "immediately adjust the spindle speed N to the preset safe speed N_safe, and adjust the feed rate F to the preset safe feed F_safe". These safety parameters are designed to quickly suppress abnormal stress during machining.

[0155] S1073: Command sending and execution.

[0156] For processing parameter optimization suggestions generated under Level 1 warning, the central processing unit sends them to the monitoring interface of the digital twin model built by S106 via a graphical user interface application (such as WebSocket or shared memory). The digital twin monitoring interface displays the suggestion instruction next to the 3D heatmap as a floating text window or a highlighted icon (alarm indicator) for operators to refer to and make decisions.

[0157] For automatic adjustment of machining parameters commands generated under Level 2 early warning, the central processing unit directly writes the adjusted safety process parameters (such as N_safe, F_safe) into the control unit of the CNC machining system via the CNC system communication interface (such as based on EtherCAT or a dedicated CNC system API). Upon receiving the command, the CNC machining system immediately interrupts the currently executing machining program segment and continues or pauses machining according to the new safety parameters, thereby achieving direct closed-loop control of the machining process.

[0158] This step is a crucial closed-loop link in realizing intelligent early warning from state perception to proactive intervention. Its core function is to transform quantitative risk assessment results into specific, tiered operational instructions. Its effects are mainly reflected in three aspects: First, by introducing a geometric complexity factor G, a stage risk base value coefficient K_phase, and characteristic fluctuation σ to dynamically calculate the early warning threshold, the early warning triggering conditions can adapt to the stability of different parts, different process stages, and different operating conditions, significantly reducing false alarms and missed alarms compared to fixed threshold methods. Second, by adopting a two-level early warning mechanism (suggestive optimization and mandatory intervention), the early warning response is graded and precise. It can provide optimization guidance in the early stages of risk to prevent the problem from worsening, and take decisive measures to ensure safety when the risk suddenly escalates. Finally, by sending instructions to the digital twin interface (human-machine interaction) and the CNC system (automatic control) respectively, a hybrid enhanced decision-making model of human-machine collaboration is constructed. This respects the operator's experience and judgment, and ensures the timeliness and automation level of the system response in emergency situations, thereby comprehensively improving the autonomy, controllability, and safety of the processing process.

[0159] This embodiment provides a self-sensing and intelligent early warning method for stress in CNC-machined aluminum alloy shell parts. The method involves distributively embedding micro-piezoelectric ceramic units on the surface of the aluminum alloy shell part to construct a conformal self-sensing array, generating piezoelectric charge signals. These signals are then transmitted via RFID backscatter modulation, conveying stress-information-containing self-sensing signals. The method receives these signals and simultaneously acquires real-time machining parameters from the CNC machining system. An adaptive weight allocation model dynamically adjusts the fusion weight coefficients according to the current machining stage, generating a real-time fusion feature vector. A pre-trained pulse neural network model is input, and a stress risk probability value is output based on a sparse pulse event-driven mechanism. A digital twin model is constructed, synchronizing the stress risk probability value with real-time machining parameters for a three-dimensional heatmap visualization. Based on the stress risk probability value and a dynamically calculated early warning threshold adapted to the part's geometric complexity, a graded early warning or machining control command is generated and executed. This achieves real-time, passive, distributed sensing and dynamic risk assessment of machining stress status, improving the timeliness of early warnings and the autonomous controllability of machining.

[0160] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only.

[0161] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for self-sensing and intelligent early warning of stress in CNC-machined aluminum alloy shell parts, characterized in that, The method includes: Multiple micro piezoelectric ceramic units are distributedly embedded on the surface of the aluminum alloy shell part to form a self-sensing array conforming to the surface of the aluminum alloy shell part. The micro piezoelectric ceramic units are used to generate piezoelectric charge signals based on the stress during CNC machining. The piezoelectric charge signal is used as a modulation source, and the modulation source is modulated by radio frequency identification backscattering to obtain a reflected radio frequency signal. The reflected radio frequency signal is then sent as a self-sensing signal containing stress information. The system receives the self-sensing signal and synchronously acquires the real-time machining parameters of the CNC machining system; wherein the CNC machining system is used to indicate the CNC system controlling the CNC machining process. The self-sensing signal and the real-time machining parameters are fused using an adaptive weight allocation model to generate a real-time fused feature vector. The adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process. The real-time fused feature vector is input into a pre-trained spiking neural network model. The spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value that characterizes the stress risk level of the current aluminum alloy shell part. A digital twin model of the aluminum alloy shell part is constructed, and the stress risk probability value and the real-time processing parameters are synchronized to the digital twin model. A three-dimensional thermal map is visualized on the three-dimensional model of the aluminum alloy shell part corresponding to the digital twin model based on the stress risk probability value. Based on the stress risk probability value and dynamic early warning threshold, a graded early warning instruction or a machining control instruction is generated and executed; wherein, the dynamic early warning threshold is determined based on the geometric complexity of the aluminum alloy shell part, the current machining stage, and the adaptive calculation of the real-time fused feature vector, and the graded early warning instruction includes at least triggering the alarm flag of the digital twin model, generating machining parameter optimization suggestions, and automatically adjusting the real-time machining parameters of the CNC machining system.

2. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The method of distributively embedding multiple micro piezoelectric ceramic units on the surface of the aluminum alloy housing part to form a self-sensing array conforming to the surface of the aluminum alloy housing part includes: In the non-assembly surface area of ​​the aluminum alloy shell part, micro-dimples matching the shape of the micro piezoelectric ceramic unit are prepared by laser micromachining; The micro piezoelectric ceramic unit is positioned in the micro recess and filled and encapsulated with high-temperature resistant insulating adhesive, so that the self-sensing array conforms to the surface of the aluminum alloy housing part and is insulated from it.

3. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of using the piezoelectric charge signal as a modulation source, modulating the modulation source through radio frequency identification backscattering to obtain a reflected radio frequency signal, and transmitting the reflected radio frequency signal as a self-sensing signal containing stress information includes: The piezoelectric charge signal is input to a load modulation circuit and converted into a digital control signal for modulating the reflection amplitude. The load modulation circuit changes the load impedance of a passive RFID tag antenna according to the digital control signal, modulates the amplitude of the incident RF carrier from the RF reader, generates the reflected RF signal, and transmits the reflected RF signal as a self-sensing signal containing stress information.

4. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of receiving the self-sensing signal and simultaneously acquiring the real-time machining parameters of the CNC machining system includes: The self-sensing signal is received by an RF reader and demodulated to obtain a stress data packet; The real-time machining parameters, including spindle speed, feed rate, and tool position, can be read in real time through the data interface of the CNC machining system. The stress data package and the real-time processing parameters are aligned and encapsulated based on a unified timestamp.

5. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of fusing the self-perceived signal and the real-time processing parameters through an adaptive weight allocation model to generate a real-time fused feature vector includes: The amplitude of the self-sensing signal is normalized to obtain normalized stress characteristics; The real-time processing parameters are numerically standardized to obtain standardized process characteristics; The normalized stress features and the standardized process features are weighted and concatenated according to dynamic weighting coefficients to generate the real-time fused feature vector.

6. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The adaptive weight allocation model dynamically adjusts the fusion weight coefficients of the self-sensing signal and the real-time machining parameters according to the current machining stage of the CNC machining process, including: Based on the process code in the real-time processing parameters, identify and output the current processing stage identifier of the CNC machining process; According to the predefined weight mapping table, query the first weight coefficient of the self-sensing signal and the second weight coefficient of the real-time processing parameter corresponding to the current processing stage identifier.

7. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of inputting the real-time fused feature vector into a pre-trained spiking neural network model, wherein the spiking neural network model processes the real-time fused feature vector based on a sparse pulse event-driven mechanism and outputs a stress risk probability value characterizing the current stress risk level of the aluminum alloy shell part, includes: The hidden layer neurons of the spiking neural network model perform membrane potential integration based on the input real-time fused feature vector. When the membrane potential exceeds the firing threshold, a pulse event is triggered and the membrane potential is reset. The output layer neurons of the spiking neural network model receive and accumulate the spiking events triggered by the hidden layer neurons; Based on the accumulated values ​​of the output layer neurons, the stress risk probability value is calculated and output through a normalization function.

8. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The process of constructing a digital twin model of the aluminum alloy shell part and synchronizing the stress risk probability value and the real-time machining parameters to the digital twin model includes: A three-dimensional geometric model and a physical property model of the aluminum alloy shell part are established to construct the digital twin model; The stress risk probability value is used as a state attribute, and the real-time processing parameters are used as a process attribute. These parameters are then updated in real time to the data nodes corresponding to the digital twin model via a communication interface.

9. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of visualizing a three-dimensional thermal map on the three-dimensional model of the aluminum alloy shell part corresponding to the digital twin model based on the stress risk probability value includes: According to the predefined color mapping rules, the stress risk probability value is mapped to the corresponding color value; The color value is rendered to the corresponding spatial position of the three-dimensional model of the aluminum alloy shell part in the digital twin model, generating the dynamically updated three-dimensional heat map.

10. The method for self-sensing and intelligent early warning of stress in CNC machined aluminum alloy shell parts according to claim 1, characterized in that, The step of generating and executing graded early warning instructions or processing control instructions based on the stress risk probability value and dynamic early warning threshold includes: The stress risk probability value is compared with a first warning threshold and a second warning threshold, wherein the second warning threshold is greater than the first warning threshold. If the stress risk probability value exceeds the first warning threshold but does not exceed the second warning threshold, an instruction to optimize the machining parameters is generated; if the stress risk probability value exceeds the second warning threshold, an instruction to automatically adjust the real-time machining parameters of the CNC machining system is generated. The generated instructions are sent to the corresponding human-machine interface or the CNC machining system for execution.

Images