Scribing saw information centralized control system

Abstract

The invention relates to the technical field of process communication and network, in particular to a scribing machine information centralized control system. Comprising a data acquisition unit, a digital signal processing unit, a motion control unit and a communication unit, and the data acquisition unit is used for acquiring key signals during the operation of the scribing machine, converting analog signals into digital signals through high-precision analog-to-digital conversion, and transmitting the digital signals to the digital signal processing unit. According to the invention, signal noise suppression and efficient recognition of cutter wear and abnormal vibration states are realized through an FPGA + DSP heterogeneous computing architecture in combination with a variable step size adaptive filtering algorithm and a lightweight neural network model, and multi-axis servo driving, closed-loop feedback and dynamic error compensation mechanisms are integrated, so that the precision of the system is improved. A control instruction is corrected on line by utilizing a thermal deformation and backlash error model, the scribing positioning precision and stability are effectively improved, the semiconductor processing efficiency and yield are remarkably improved, and meanwhile, a reusable technical normal form is provided for intelligent and networked upgrading of high-end manufacturing equipment.

Description

Technical Field

[0001] The present invention relates to the field of process communication and network technology, and in particular to a dicing machine information centralized control system. Background Art

[0002] Dicing machines are key equipment in precision processing fields such as semiconductor manufacturing and electronic packaging. They are mainly used to cut entire semiconductor wafers, electronic substrates (such as PCB boards) or other brittle materials (such as glass and ceramics) into individual chips, devices or units. Their core function is to achieve precise material separation with micron-level accuracy through cutting methods such as high-speed rotating blades, laser beams or plasma, while ensuring the quality of the cut edges (such as flatness and crack control) to avoid damage to internal circuits or structures. Dicing machines are widely used in production links such as integrated circuits, discrete devices, and advanced packaging. They are important equipment for the key process of "wafer manufacturing → packaging and testing → finished products" in the semiconductor industry chain. With the continuous advancement of Moore's Law and the popularization of advanced packaging technology, the cutting accuracy requirements for semiconductor materials have been upgraded from the early micron level to the submicron level, and higher requirements have been placed on cutting efficiency, yield rate and process compatibility.

[0003] In existing technologies, traditional dicing machine control systems mainly rely on analog signal processing circuits and a single closed-loop control architecture, exposing the following bottlenecks:

[0004] Insufficient real-time signal processing: The fixed parameter filtering algorithm cannot dynamically adapt to sudden changes in signal noise caused by tool wear, material hardness changes, etc. during the cutting process, resulting in delayed extraction of fault features such as tool breakage and spindle abnormalities. The fault response time is greater than 100ms, which can easily lead to the scrapping of batch products.

[0005] Multi-source error compensation is not accurate: it can only achieve static compensation for a single mechanical error, and lacks the ability to correct dynamic errors such as thermal deformation and vibration coupling errors in real time. The cumulative positioning error can reach ±5μm, which exceeds the accuracy requirements of advanced packaging.

[0006] The system's communication reliability is insufficient and its intelligent collaboration capabilities are weak: it uses a non-standardized communication protocol, and the data interaction delay with the factory's MES system and visual inspection equipment is greater than 100ms. In addition, there is a lack of a priority scheduling mechanism, and the failure rate of emergency control command transmission is as high as 5%. At the same time, it relies on manual experience to adjust process parameters, and cannot realize intelligent functions such as tool life prediction and cutting path adaptive optimization, resulting in low production efficiency.

[0007] Based on this, the present invention provides a centralized information control system for a dicing machine to solve the technical problems raised above. Summary of the Invention

[0008] The purpose of the present invention is to provide a centralized information control system for a dicing machine. The FPGA+DSP heterogeneous computing architecture of the present invention is combined with a variable step-size adaptive filtering algorithm and a lightweight neural network model to achieve signal noise suppression and efficient identification of tool wear and abnormal vibration states, and integrate multi-axis servo drive, closed-loop feedback and dynamic error compensation mechanism, and use thermal deformation and tooth gap error models to correct control instructions online, effectively improving the dicing positioning accuracy and stability, significantly improving the efficiency and yield of semiconductor processing, and at the same time providing a reusable technical paradigm for the intelligent and networked upgrade of high-end manufacturing equipment.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] The present invention provides a centralized information control system for a dicing machine, comprising a data acquisition unit, a digital signal processing unit, a motion control unit, and a communication unit, wherein:

[0011] The data acquisition unit is used to collect key signals during the operation of the dicing machine, convert the analog signals into digital signals through high-precision analog-to-digital conversion, and then transmit them to the digital signal processing unit;

[0012] The digital signal processing unit is used to realize real-time signal processing and feature extraction based on a variable step-size adaptive filtering algorithm and a lightweight neural network model, and to perform filtering and intelligent decision-making in parallel through an FPGA+DSP heterogeneous computing architecture, thereby dynamically optimizing control parameters;

[0013] The motion control unit is used to perform high-precision trajectory tracking, position closed-loop control and dynamic error compensation based on the state recognition results and control instructions output by the digital signal processing unit, combined with the multi-axis servo drive and redundant communication mechanism;

[0014] The communication unit is used to support communication with external systems based on the industrial Ethernet protocol and transmit data according to priority.

[0015] The data acquisition unit includes a signal sensing module, a signal conditioning module, an analog-to-digital conversion module, and a data buffer module, wherein:

[0016] The signal sensing module is used to deploy multiple types of sensors such as vibration, current, and temperature to obtain physical signals from the dicing machine;

[0017] The signal conditioning module is used to pre-process the original analog signal by amplifying and filtering;

[0018] The analog-to-digital conversion module is used to convert the conditioned analog signal into a digital signal through a high-precision ADC;

[0019] The data buffer module is used to temporarily store digital signals and transmit them to the digital signal processing unit in a time sequence.

[0020] The vibration sensor is arranged at the spindle bearing position of the dicing machine and is used to collect vibration signals during the cutting process; the current sensor is connected to the servo motor drive circuit and is used to collect the motor running current signal; there are three temperature sensors, one of which is arranged on the outside of the spindle bearing seat and adopts a Pt100 platinum resistance temperature sensor to monitor the temperature changes of the spindle bearing in real time; the other is embedded in the servo motor winding and adopts a K-type thermocouple to measure the actual working temperature of the motor winding; the last one is arranged near the coolant nozzle in the cutting area and adopts an infrared temperature sensor to detect the temperature of the cutting point in a non-contact manner.

[0021] The digital signal processing unit includes an adaptive filtering processing module, a feature extraction module, a decision control module, and a heterogeneous computing module, wherein:

[0022] The adaptive filtering processing module is used to filter out signal noise in real time based on a variable step size algorithm;

[0023] The feature extraction module is used to extract key features of tool wear, abnormal vibration, and sudden change of motor load from the signal through a lightweight neural network;

[0024] The decision control module is used to dynamically generate control parameters according to the feature analysis results;

[0025] The heterogeneous computing module is used to accelerate filtering operations through FPGA and perform model reasoning through DSP.

[0026] The feature extraction process of the lightweight neural network model is specifically performed as follows:

[0027] A1: Feature Map: The feature map is calculated by the depth-wise separable convolution layer, and the formula is F(x ' )=W dw *(W pw *x ' +b pw )+b dw , where W dw , W pw are depth convolution kernel and planar convolution kernel respectively, b pw , b dw is the bias term;

[0028] A2: Weighted attention mechanism: Calculating channel importance based on global average pooling Where N is the feature map size and σ is the activation function;

[0029] A3: Feature fusion: add feature maps of different sizes according to weights Where M is the number of feature maps;

[0030] A4: Classification and recognition: through the fully connected layer and Softmax function The probability distribution y of each category is calculated to identify the status of tool wear, abnormal vibration, and sudden change of motor load.

[0031] The heterogeneous computing module is used to accelerate filtering operations through FPGA and perform model reasoning through DSP. The specific operations are as follows:

[0032] B1: The raw signal output by the data acquisition unit is synchronously transmitted to the FPGA and DSP: the FPGA receives high-frequency real-time signals such as vibration and current and directly performs variable-step adaptive filtering; the DSP receives pre-processed feature data and performs lightweight neural network inference;

[0033] B2: FPGA uses a hardware pipeline architecture to execute filtering algorithms on multiple signals in parallel, with a per-channel filtering delay of ≤1μs. DSP uses a multi-core processor to perform batch inference on feature vectors, with a single inference time of ≤10μs.

[0034] B3: The filtered signal output by the FPGA and the characteristic analysis results output by the DSP interact in real time through a high-speed bus to generate the final control parameters;

[0035] B4: When an abnormal signal is detected, computing resource allocation is automatically adjusted: the FPGA increases the filtering priority to the highest level to ensure that abnormal signals are processed first; the DSP suspends non-critical tasks and concentrates computing power on executing the fault feature recognition algorithm.

[0036] The motion control unit includes a multi-axis drive module, a closed-loop feedback module, an error compensation module, and a redundant communication module, wherein:

[0037] The multi-axis drive module is used to control the start and stop and speed of the X, Y, Z and rotary axis servo motors;

[0038] The closed-loop feedback module is used to correct the position error in real time through encoder and grating ruler feedback;

[0039] The error compensation module is used to calculate the mechanical error in real time and dynamically correct the control instructions;

[0040] The redundant communication module is used to ensure reliable transmission of control instructions through a dual industrial Ethernet channel.

[0041] The error compensation module realizes dynamic compensation of mechanical errors. The specific operations are as follows:

[0042] C1: real-time acquisition of encoder feedback signals and temperature sensor data;

[0043] C2: Based on the preset thermal deformation error model and backlash error model, calculate the current axial thermal deformation and backlash error values respectively;

[0044] C3: Generate the total error compensation by adding up the error terms;

[0045] C4: Dynamically add the total error compensation to the position control instruction to correct the output pulse or voltage signal.

[0046] The specific expression of the thermal deformation error model is:

[0047] Δx thermal =α·L·(T(t)-T0)

[0048] Where α is the linear expansion coefficient of the material, L is the length of the guide rail, T(t) is the current temperature, and T0 is the reference temperature.

[0049] The priority transmission mechanism of the communication unit includes the following:

[0050] Categorize data into three priority levels:

[0051] Level 1 priority: contains emergency control instructions and fault alarm information, with a transmission delay of no more than 1ms;

[0052] Level 2 priority: contains real-time status data such as position and speed feedback, with a transmission cycle of no more than 10ms;

[0053] Level 3 priority: Contains historical data and non-real-time configuration information, and is transmitted using time-division multiplexing.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention uses FPGA+DSP heterogeneous computing architecture, combined with variable step-size adaptive filtering algorithm and lightweight neural network model, to achieve signal noise suppression and efficient identification of tool wear and abnormal vibration status, and integrates multi-axis servo drive, closed-loop feedback and dynamic error compensation mechanism, and uses thermal deformation and tooth gap error model to correct control instructions online, effectively improving the dicing positioning accuracy and stability, significantly improving the efficiency and yield of semiconductor processing, and at the same time providing a reusable technical paradigm for the intelligent and networked upgrade of high-end manufacturing equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a system diagram of a centralized information control system for a dicing machine according to the present invention.

[0057] Figure 2 The present invention provides a flow chart of communication priority scheduling in a centralized information control system for a dicing machine.

[0058] Figure 3 This is a closed-loop flow chart of motion control in a centralized information control system for a dicing machine according to the present invention.

[0059] Description of Figure Numbers:

[0060] 100. Data acquisition unit; 101. Signal sensing module; 102. Signal conditioning module; 103. Analog-to-digital conversion module; 104. Data buffer module; 200. Digital signal processing unit; 201. Adaptive filtering processing module; 202. Feature extraction module; 203. Decision control module; 204. Heterogeneous computing module; 300. Motion control unit; 301. Multi-axis drive module; 302. Closed-loop feedback module; 303. Error compensation module; 304. Redundant communication module; 400. Communication unit. DETAILED DESCRIPTION

[0061] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example:

[0063] like Figure 1-Figure 3 As shown, this embodiment provides a centralized information control system for a dicing machine, including a data acquisition unit 100, a digital signal processing unit 200, a motion control unit 300, and a communication unit 400, wherein: the data acquisition unit 100: is used to collect key signals during the operation of the dicing machine, and converts the analog signals into digital signals through high-precision analog-to-digital conversion and then transmits them to the digital signal processing unit 200; the digital signal processing unit 200: is used to realize real-time signal processing and feature extraction based on a variable step-size adaptive filtering algorithm and a lightweight neural network model, and performs filtering and intelligent decision-making in parallel through an FPGA+DSP heterogeneous computing architecture, and dynamically optimizes control parameters; the motion control unit 300: is used to perform high-precision trajectory tracking, position closed-loop control and dynamic error compensation according to the state recognition results and control instructions output by the digital signal processing unit 200, combined with multi-axis servo drive and redundant communication mechanism; the communication unit 400: is used to support communication with external systems based on the industrial Ethernet protocol, and transmit data according to priority.

[0064] In this embodiment, it should be noted that: the data acquisition unit 100 collects signals in real time and transmits them to the digital signal processing unit 200 for intelligent analysis and decision-making, and the generated optimization instructions interact with the motion control unit 300 through the communication unit 400 to perform high-precision motion, and at the same time, the motion state is fed back to the data acquisition unit 100 to realize closed-loop control.

[0065] In the present invention, the data acquisition unit 100 includes a signal sensing module 101, a signal conditioning module 102, an analog-to-digital conversion module 103, and a data buffer module 104, wherein: the signal sensing module 101 is used to deploy multiple types of sensors such as vibration, current, and temperature to obtain the physical signal of the dicing machine; the vibration sensor is arranged at the spindle bearing position of the dicing machine, and is used to collect vibration signals during the cutting process; the current sensor is connected to the servo motor drive circuit, and is used to collect the motor running current signal; there are three temperature sensors, one of which is arranged on the outside of the spindle bearing seat, and adopts a Pt100 platinum resistance temperature sensor, and is used to monitor the temperature changes of the spindle bearing in real time; the other is pre-buried in the servo motor winding, and adopts a K-type thermocouple, and is used to measure the actual working temperature of the motor winding; the last one is arranged near the coolant nozzle in the cutting area, and adopts an infrared temperature sensor, and is used to non-contact detect the temperature of the cutting point. The signal conditioning module 102 is used to pre-process the original analog signal by amplifying and filtering; the analog-to-digital conversion module 103 is used to convert the conditioned analog signal into a digital signal through a high-precision ADC; the data buffer module 104 is used to temporarily store the digital signal and package it in a time sequence for transmission to the digital signal processing unit 200.

[0066] In this embodiment, it should be noted that the vibration, current, and temperature signals collected by the signal sensing module 101 are pre-processed by the signal conditioning module 102, converted into digital signals by the analog-to-digital conversion module 103, and finally packaged and transmitted to the digital signal processing unit 200 in a time sequence through the data buffer module 104.

[0067] In addition, it should be noted that the vibration sensor is deployed on the spindle bearing seat, with a sampling frequency of ≥10kHz, capable of capturing abnormal vibration frequencies between 20 and 2000Hz. The current sensor uses a closed-loop Hall effect device (accuracy ±0.5%) to monitor the servo motor's three-phase current waveform in real time. The temperature sensor array covers the spindle, motor windings, and cutting area, with a temperature resolution of 0.1°C. The high-precision analog-to-digital conversion uses a 24-bit Σ-Δ ADC, supporting 8-channel simultaneous sampling at a sampling rate of no less than 100kHz to ensure undistorted signals. Signal conditioning module 102 uses a second-order Butterworth low-pass filter (cutoff frequency 500Hz) to suppress high-frequency noise, and amplifies the signal to the standard range of [-5V, +5V] via an instrumentation amplifier.

[0068] In the present invention, the digital signal processing unit 200 includes an adaptive filtering processing module 201, a feature extraction module 202, a decision control module 203, and a heterogeneous computing module 204, wherein: the adaptive filtering processing module 201 is used to filter out signal noise in real time based on a variable step size algorithm; the feature extraction module 202 is used to extract key features of tool wear, vibration abnormality, and motor load mutation from the signal through a lightweight neural network; the feature extraction process of the lightweight neural network model is specifically performed as follows: A1: Feature mapping: The feature map is calculated through a depth-separable convolution layer, and the formula is F(x ' )=W dw *(W pw *x ' +b pw )+b dw , where W dw , W pw are depth convolution kernel and planar convolution kernel respectively, b pw , b dw is the bias term; A2: Attention mechanism weighting: channel importance is calculated based on global average pooling Where N is the feature map size, σ is the activation function; A3: Feature fusion: add feature maps of different sizes according to the weights Where M is the number of feature maps; A4: Classification recognition: through the fully connected layer and Softmax function The probability distribution y for each category is calculated to identify states such as tool wear, abnormal vibration, and sudden changes in motor load. The decision-making control module 203 dynamically generates control parameters based on feature analysis results. The heterogeneous computing module 204 uses the FPGA to accelerate filtering operations and the DSP to perform model inference. The specific operations are as follows: B1: The raw signal output by the data acquisition unit 100 is synchronously transmitted to the FPGA and DSP. The FPGA receives high-frequency, real-time signals such as vibration and current and directly performs variable-step adaptive filtering. The DSP receives preprocessed feature data and performs lightweight neural network inference. B2: The FPGA uses a hardware pipeline architecture to parallelize filtering algorithms on multiple signals, for example, with a per-channel filtering delay of ≤1μs. The DSP utilizes a multi-core processor to perform batch inference on feature vectors, for example, with a single inference time of ≤10μs. B3: The filtered signal output by the FPGA and the feature analysis results output by the DSP interact in real time via a high-speed bus to generate the final control parameters. B4: When an abnormal signal is detected, computing resource allocation is automatically adjusted. The FPGA increases filtering priority to the highest level to ensure that the abnormal signal is processed first. The DSP suspends non-critical tasks and concentrates computing power on executing the fault feature identification algorithm.

[0069] In this embodiment, it should be noted that: the heterogeneous computing module 204 coordinates the parallel processing of the FPGA and DSP: the signal after real-time noise filtering by the adaptive filtering processing module 201 and the neural network analysis results of the feature extraction module 202 are integrated in the decision control module 203 to generate optimization instructions.

[0070] In addition, it should be noted that the variable step size adaptive filtering algorithm adopts the normalized least mean square algorithm, and the step size factor μ is dynamically adjusted between 0.01 and 0.9 according to the signal-to-noise ratio. The specific formula of the variable step size algorithm is: The lightweight neural network model is a deep separable convolutional neural network (DS-CNN), with no more than 500,000 parameters and optimized using INT8 quantization technology. The DSP uses the TI C6678 multi-core processor, and the neural network inference uses quantized INT8 precision, with a single inference time of ≤5ms.

[0071] In the present invention, the motion control unit 300 includes a multi-axis drive module 301, a closed-loop feedback module 302, an error compensation module 303, and a redundant communication module 304, wherein: the multi-axis drive module 301 is used to control the start and stop and speed of the X, Y, Z and rotary axis servo motors; the closed-loop feedback module 302 is used to correct the position error in real time through encoder and grating scale feedback; the error compensation module 303 is used to calculate the mechanical error in real time and dynamically correct the control instructions; the specific operations are as follows: C1: real-time acquisition of encoder feedback signals and temperature sensor data; C2: based on the preset thermal deformation error model and backlash error model, respectively calculate the current axial thermal deformation and backlash error value; the specific expression of the thermal deformation error model is:

[0072] Δx thermal =α·L·(T(t)-T0)

[0073] Where α is the material linear expansion coefficient, L is the guide rail length, T(t) is the current temperature, and T0 is the reference temperature. C3: The error terms are summed to generate a total error compensation. C4: The total error compensation is dynamically added to the position control command to correct the output pulse or voltage signal. Redundant communication module 304 ensures reliable transmission of control commands via dual industrial Ethernet channels.

[0074] In this embodiment, it should be noted that: the multi-axis drive module 301 executes motion instructions, the closed-loop feedback module 302 monitors position deviations in real time, the error compensation module 303 dynamically corrects mechanical errors, and the redundant communication module 304 ensures reliable transmission of the control link.

[0075] In addition, it should be noted that the redundant communication mechanism includes a master-slave dual-channel industrial Ethernet channel, with the master channel using the Profinet protocol and the slave channel using the EtherCAT protocol. If the master channel communication is interrupted, the slave channel automatically takes over control command transmission, with a switchover time of no more than 2ms. The multi-axis servo drive adopts a three-loop control architecture, including position, velocity, and current loops. The position loop control period is no more than 100μs, the velocity loop period is 50μs, and the current loop period is 20μs. Backlash error compensation uses a combination of preload and backlash compensation tables to dynamically correct step errors through a lookup table, with a compensation accuracy of ≤±0.2μm.

[0076] In the present invention, the priority transmission mechanism of the communication unit 400 includes the following: dividing data into three priority levels: Level 1 priority: including emergency control instructions and fault alarm information, with a transmission delay of no more than 1ms; Level 2 priority: including real-time status data such as position and speed feedback, with a transmission cycle of no more than 10ms; Level 3 priority: including historical data and non-real-time configuration information, and transmitted in a time-division multiplexing manner.

[0077] In this embodiment, it should be noted that the communication unit 400 implements intelligent data scheduling through a three-level priority transmission mechanism: the first-level emergency instruction ≤1ms prioritizes bandwidth, the second-level real-time status data ≤10ms ensures basic control, and the third-level non-real-time data is time-division multiplexed.

[0078] In addition, it should be noted that: Priority scheduling: Based on IEEE 802.1Qbv time-sensitive networking technology, dedicated time slots are allocated for urgent instructions.

[0079] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0080] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A centralized information control system for a dicing machine, characterized in that: The system comprises a data acquisition unit (100), a digital signal processing unit (200), a motion control unit (300), and a communication unit (400), wherein: The data acquisition unit (100) is used to collect key signals during the operation of the dicing machine, convert the analog signals into digital signals through high-precision analog-to-digital conversion, and then transmit them to the digital signal processing unit (200); The digital signal processing unit (200) is used to realize real-time signal processing and feature extraction based on a variable step-size adaptive filtering algorithm and a lightweight neural network model, and to perform filtering and intelligent decision-making in parallel through an FPGA+DSP heterogeneous computing architecture, thereby dynamically optimizing control parameters; The motion control unit (300) is used to perform high-precision trajectory tracking, position closed-loop control and dynamic error compensation based on the state recognition result and control instruction output by the digital signal processing unit (200) in combination with multi-axis servo drive and redundant communication mechanism; The communication unit (400) is used to support communication with an external system based on the industrial Ethernet protocol and transmit data according to priority.

2. A dicing machine information centralized control system according to claim 1, characterized in that: The data acquisition unit (100) comprises a signal sensing module (101), a signal conditioning module (102), an analog-to-digital conversion module (103), and a data buffer module (104), wherein: The signal sensing module (101) is used to deploy multiple types of sensors such as vibration, current, and temperature to obtain physical signals from the dicing machine; The signal conditioning module (102) is used for pre-processing the original analog signal by amplifying and filtering; The analog-to-digital conversion module (103) is used to convert the conditioned analog signal into a digital signal through a high-precision ADC; The data buffer module (104) is used to temporarily store digital signals and package and transmit them to the digital signal processing unit (200) in a time sequence.

3. A dicing machine information centralized control system according to claim 2, characterized in that: The vibration sensor is arranged at the spindle bearing position of the dicing machine and is used to collect vibration signals during the cutting process; the current sensor is connected to the servo motor drive circuit and is used to collect the motor running current signal; there are three temperature sensors, one of which is arranged on the outside of the spindle bearing seat and adopts a Pt100 platinum resistance temperature sensor to monitor the temperature changes of the spindle bearing in real time; the other is embedded in the servo motor winding and adopts a K-type thermocouple to measure the actual working temperature of the motor winding; the last one is arranged near the coolant nozzle in the cutting area and adopts an infrared temperature sensor to detect the temperature of the cutting point in a non-contact manner.

4. A dicing machine information centralized control system according to claim 1, characterized in that: The digital signal processing unit (200) comprises an adaptive filtering processing module (201), a feature extraction module (202), a decision control module (203), and a heterogeneous computing module (204), wherein: The adaptive filtering processing module (201) is used to filter out signal noise in real time based on a variable step size algorithm; The feature extraction module (202) is used to extract key features of tool wear, abnormal vibration, and sudden change of motor load from the signal through a lightweight neural network; The decision control module (203) is used to dynamically generate control parameters according to the feature analysis results; The heterogeneous computing module (204) is used to accelerate filtering operations through FPGA and perform model reasoning through DSP.

5. A dicing machine information centralized control system according to claim 4, characterized in that: The feature extraction process of the lightweight neural network model is specifically performed as follows: A1: Feature Map: The feature map is calculated by the depth-wise separable convolution layer, and the formula is F(x ' )=W dw *(W pw *x ' +b pw )+b dw , where W dw , W pw are depth convolution kernel and planar convolution kernel respectively, b pw , b dw is the bias term; A2: Weighted attention mechanism: Calculating channel importance based on global average pooling Where N is the feature map size and σ is the activation function; A3: Feature fusion: add feature maps of different sizes according to weights Where M is the number of feature maps; A4: Classification and recognition: through the fully connected layer and Softmax function The probability distribution y of each category is calculated to identify the status of tool wear, abnormal vibration, and sudden change of motor load.

6. A dicing machine information centralized control system according to claim 4, characterized in that: The heterogeneous computing module (204) is used to accelerate filtering operations through FPGA and perform model reasoning through DSP. The specific operations are as follows: B1: The original signal output by the data acquisition unit (100) is synchronously transmitted to the FPGA and DSP: the FPGA receives high-frequency real-time signals such as vibration and current, and directly performs variable step-size adaptive filtering; the DSP receives pre-processed feature data and performs lightweight neural network inference; B2: FPGA uses a hardware pipeline architecture to execute filtering algorithms on multiple signals in parallel, with a per-channel filtering delay of ≤1μs. DSP uses a multi-core processor to perform batch inference on feature vectors, with a single inference time of ≤10μs. B3: The filtered signal output by the FPGA and the characteristic analysis results output by the DSP interact in real time through a high-speed bus to generate the final control parameters; B4: When an abnormal signal is detected, computing resource allocation is automatically adjusted: the FPGA increases the filtering priority to the highest level to ensure that abnormal signals are processed first; the DSP suspends non-critical tasks and concentrates computing power on executing the fault feature recognition algorithm.

7. The centralized information control system for dicing machines according to claim 1, characterized in that: The motion control unit (300) comprises a multi-axis drive module (301), a closed-loop feedback module (302), an error compensation module (303), and a redundant communication module (304), wherein: The multi-axis drive module (301) is used to control the start and stop and speed of the X, Y, Z and rotary axis servo motors; The closed-loop feedback module (302) is used to correct the position error in real time through encoder and grating ruler feedback; The error compensation module (303) is used to calculate the mechanical error in real time and dynamically correct the control instruction; The redundant communication module (304) is used to ensure reliable transmission of control instructions through a dual-path industrial Ethernet channel.

8. The centralized information control system for dicing machines according to claim 7, characterized in that: The error compensation module (303) realizes dynamic compensation of mechanical errors, and the specific operations are as follows: C1: real-time acquisition of encoder feedback signals and temperature sensor data; C2: Based on the preset thermal deformation error model and backlash error model, calculate the current axial thermal deformation and backlash error values respectively; C3: Generate the total error compensation by adding up the error terms; C4: Dynamically add the total error compensation to the position control instruction to correct the output pulse or voltage signal.

9. The centralized information control system for dicing machines according to claim 8, characterized in that: The specific expression of the thermal deformation error model is: Δx thermal =α·L·(T(t)-T0) Where α is the linear expansion coefficient of the material, L is the length of the guide rail, T(t) is the current temperature, and T0 is the reference temperature.

10. The centralized information control system for dicing machines according to claim 1, characterized in that: The priority transmission mechanism of the communication unit (400) includes the following: Categorize data into three priority levels: Level 1 priority: contains emergency control instructions and fault alarm information, with a transmission delay of no more than 1ms; Level 2 priority: contains real-time status data such as position and speed feedback, with a transmission cycle of no more than 10ms; Level 3 priority: Contains historical data and non-real-time configuration information, and is transmitted using time-division multiplexing.

Images