Numerical control lathe multi-sensor fusion monitoring system based on edge calculation

Abstract

The invention relates to the technical field of sensing equipment, in particular to a numerical control lathe multi-sensor fusion monitoring system based on edge calculation. Comprising edge computing nodes deployed in a numerical control system control cabinet, a hardware platform at least comprising a quad-core ARM Cortex-A72 processor, a 16GB DDR4 memory and an NVMe solid state disk is configured, and a real-time operating system is operated; the sensor network layer is a multi-mode sensing unit composed of a vibration sensor array, a temperature sensor group, a current sensor and an acoustic emission sensor, a vibration sensor adopts a three-way accelerometer to be arranged on the surface of a main shaft bearing seat, and the temperature sensor adopts a PT100 platinum resistor to be attached to a main shaft motor shell; the data fusion layer is used for implementing a dynamic weight distribution algorithm and automatically adjusting a fusion weight coefficient of data of each sensor according to the current processing stage; according to the scheme, the technical normal form of cloud dependence can be broken through, and an edge-end multi-sensor fusion architecture with real-time decision-making capability is constructed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sensing equipment, in particular to a multi-sensor fusion monitoring system for a numerical control lathe based on edge computing. BACKGROUND

[0002] In the field of high-end equipment manufacturing, numerical control lathes as mother equipment directly affect the machining quality of strategic industries such as aerospace and precision molds. The existing numerical control lathe monitoring system generally adopts a sensor collection-cloud computing architecture, and data needs to be transmitted to a remote server for fusion analysis through an industrial Ethernet, resulting in an end-to-end time delay generally exceeding 300 ms. This technical architecture exposes two major core defects in dynamic processing scenarios such as variable depth and variable speed: first, the fixed sampling rate mechanism causes data redundancy and key information loss to coexist. When the spindle speed suddenly changes from 1500 rpm to 3000 rpm, the existing cloud monitoring system cannot adaptively adjust the sampling rate, resulting in a large amount of invalid data occupying bandwidth, and the Nyquist sampling theorem limits the loss of chatter frequency, causing a bearing fault warning delay of up to 4.7 seconds. Second, the static weight allocation algorithm is difficult to adapt to multi-source heterogeneous data. The fixed weight fusion strategy used in the existing technology cannot dynamically adjust the fusion ratio of vibration and temperature data when the sensor baseline drifts due to spindle motor temperature rise, resulting in 7 consecutive bearing cage rupture misdiagnosis accidents in the actual production of a certain automobile parts manufacturer. The essence of this technical bottleneck is the physical isolation characteristic of the cloud architecture, which makes the real-time performance and adaptive ability of data fusion irreconcilable.

[0003] Based on the above technical problems, there is an urgent need for a technical paradigm that can break through the dependence on the cloud and build an edge multi-sensor fusion architecture with real-time decision-making capabilities. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to propose a multi-sensor fusion monitoring system for a numerical control lathe based on edge computing, comprising: Edge computing layer: an edge computing node deployed in the numerical control system control cabinet, configured with a hardware platform containing a four-core ARM Cortex-A72 processor, 16GB DDR4 memory and an NVMe solid state disk, and running a real-time operating system; Sensor network layer: a multi-modal sensing unit composed of a vibration sensor array, a temperature sensor group, a current sensor and an acoustic emission sensor, wherein the vibration sensor is a three-axis accelerometer arranged on the surface of the spindle bearing seat, and the temperature sensor is a PT100 platinum resistance attached to the spindle motor shell, the sensor network layer further comprises an adaptive sampling rate adjustment mechanism for dynamically adjusting the sampling frequency according to the signal characteristics; Data fusion layer: Implement dynamic weight distribution algorithm, automatically adjust the fusion weight coefficient of each sensor data according to the current processing stage; Abnormal decision layer: Establish a lightweight fault classification model based on support vector machine, the input dimension of the model is not more than 12; Communication layer: Adopt time sensitive network protocol to realize deterministic transmission of sensor data and control command; Cloud collaborative layer: Through 5G private network and edge node to establish two-way communication, only when potential fault is detected, cloud deep diagnosis process is triggered.

[0005] Preferably, the edge computing node is coupled with the numerical control unit of the numerical control system through a PCIe Gen3x4 interface, and the data interaction period is set to 1ms; the edge node is configured with dual redundant power supply modules, and the switching time from the main power supply to the standby power supply is ≤15ms when the main power supply fails; the edge computing node is built-in hardware security module, supporting AES-256 encryption algorithm for real-time encryption of transmission data; the node shell is made of aluminum-based composite material.

[0006] Further preferably, the three-axis accelerometer of the vibration sensor array is fixed to the main shaft bearing seat through an M5 stainless steel threaded joint; the sensor signal line adopts a twisted shielded cable, the shielding layer ground resistance is ≤0.1Ω, and the cable bending radius is ≥10 times the cable outer diameter; each accelerometer is configured with an independent charge amplifier module; a 0.5mm thick silica gel damping layer is arranged between the mounting base and the machine tool body for isolating high-frequency vibration interference.

[0007] Further preferably, the three PT100 sensors of the temperature sensor group are respectively embedded in the main shaft motor stator winding, the rotor shaft end and the front bearing outer ring; the sensor lead adopts a four-core compensation wire, and the cold end compensation is realized by a digital temperature sensor; each temperature sensor is configured with an independent multiplexing switch, and the sampling period can be programmed; the sensor shell is made of 316L stainless steel.

[0008] Further preferably, the dynamic weight distribution algorithm adopts an improved entropy weight method, and the weight coefficient calculation formula is: ; In the formula: is the dynamic weight coefficient of the jth sensor; is the information entropy, wherein k is the normalization constant; is the preset reference temperature threshold, is the real-time temperature value of the jth sensor; n is the total number of sensors, and m is the sampling window length.

[0009] Further preferably, the adaptive sampling rate adjustment mechanism is realized by the following formula: ; In the formula: is the adjusted sampling frequency; is the basic sampling rate; is the current vibration signal standard deviation, is the preset maximum allowable standard deviation; alpha is the adjustment coefficient, and beta is the exponential factor.

[0010] Further preferably, it further comprises an edge-cloud collaborative working mode, and the edge-cloud collaborative working mode adopts the following data fusion strategy: ; In the formula: is the final decision result; is the edge node preliminary diagnosis conclusion; is the cloud deep analysis result; and lambda is the confidence coefficient.

[0011] Further preferably, when the support vector machine model adopts the improved grey wolf optimization algorithm for parameter optimization, the wolf population size is set to 30-50 individuals, and the maximum iteration number is 80-120 times; the fitness function is designed as the weighted sum of the classification accuracy and the model complexity, the weight coefficient is dynamically adjusted according to the cross-validation result, the population position is initialized by introducing chaos mapping to avoid the optimization process from falling into local optimum; and the model training data set contains three types of samples of normal state, early fault and serious fault, and the number ratio of each type of sample is kept at 3:1:1.

[0012] Further preferably, the time-sensitive network configuration contains a dual-redundancy switch, and the switching time of the master and standby switches is ≤500 ns; 8 priority queues are allocated for sensor data streams, and vibration data, temperature data and current data correspond to Class 3, Class 4 and Class 5 priority respectively; the gate list configuration is synchronized with the interpolation period of the numerical control system; and the network fault recovery mechanism contains a link aggregation control protocol and a fast spanning tree protocol.

[0013] Further preferably, the 5G private network adopts a 2.6 GHz frequency band, and a distributed micro base station is deployed to realize full coverage of the factory; and the user plane function is sunk to the factory edge machine room, and connected to the core network through an N3 interface; A dual-transmission-reception transmission mechanism is configured, important control instructions are sent through two independent paths, and the receiving end selects the first-arriving data packet; The network slicing technology divides three independent logical channels, which respectively bear sensor data, control instructions and video monitoring services.

[0014] Technical effects: The overall architecture of the system includes an edge computing layer, a sensor network layer, a data fusion layer, an anomaly decision layer, a communication layer, and a cloud collaboration layer. The edge computing layer adopts a hardware configuration of a four-core ARM Cortex-A72 processor and 16GB DDR4 memory, runs a real-time operating system, and ensures low-delay data processing. The sensor network layer is composed of vibration, temperature, current, and acoustic emission sensors. The vibration sensor is laid out on the surface of the main shaft bearing seat using a three-axis accelerometer, and the temperature sensor is attached to the motor housing using a PT100 platinum resistance. The data fusion layer adjusts the sensor data weight through a dynamic weight distribution algorithm, the anomaly decision layer realizes fault classification based on a lightweight SVM model, the communication layer ensures deterministic transmission using the TSN protocol, and the cloud collaboration layer realizes two-way communication through a 5G private network. The problems of high data delay, waste of computing resources, poor adaptability to dynamic working conditions, and other issues in traditional CNC lathe monitoring systems are solved. The single sensor or static weight fusion scheme cannot meet the needs of complex processing scenarios, cloud computing dependence leads to insufficient real-time performance, and sensor data fusion lacks adaptive ability. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The block diagram of the CNC lathe multi-sensor fusion monitoring system based on edge computing of the present application. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0017] Please refer to Figure 1 The traditional CNC lathe monitoring system has problems such as high data delay, waste of computing resources, poor adaptability to dynamic working conditions, and other issues. The single sensor or static weight fusion scheme cannot meet the needs of complex processing scenarios, cloud computing dependence leads to insufficient real-time performance, and sensor data fusion lacks adaptive ability. Based on this, the present embodiment provides a CNC lathe multi-sensor fusion monitoring system based on edge computing, which comprises: Edge computing layer: edge computing nodes deployed in the CNC system control cabinet, configured with a hardware platform containing at least a four-core ARM Cortex-A72 processor, 16GB DDR4 memory and NVMe solid state disk, running a real-time operating system; Sensor network layer: a multi-modal sensing unit composed of a vibration sensor array, a temperature sensor group, a current sensor and an acoustic emission sensor, wherein the vibration sensor is laid out on the surface of the main shaft bearing seat using a three-axis accelerometer, and the temperature sensor is attached to the main shaft motor housing using a PT100 platinum resistance; Data fusion layer: Implement dynamic weight distribution algorithm, automatically adjust the fusion weight coefficient of each sensor data according to the current processing stage; Abnormal decision layer: Establish a lightweight fault classification model based on support vector machine, the input dimension of the model is not more than 12; Communication layer: Adopt time-sensitive network protocol to realize deterministic transmission of sensor data and control instructions; Cloud collaboration layer: Establish bidirectional communication with edge nodes through 5G private network, trigger cloud deep diagnosis process only when potential fault is detected.

[0018] It is worth mentioning that: this embodiment defines the overall architecture of the system, including edge computing layer, sensor network layer, data fusion layer, abnormal decision layer, communication layer and cloud collaboration layer. The edge computing layer adopts four-core ARM Cortex-A72 processor and 16GB DDR4 memory hardware configuration, runs real-time operating system, and ensures low-delay data processing. The sensor network layer is composed of vibration, temperature, current and acoustic emission sensors. The vibration sensor is laid in the main shaft bearing seat with three-axis accelerometer, and the temperature sensor uses PT100 platinum resistance attached to the motor shell. The data fusion layer adjusts the weight of sensor data through dynamic weight distribution algorithm, the abnormal decision layer realizes fault classification based on lightweight SVM model, the communication layer adopts TSN protocol to ensure deterministic transmission, and the cloud collaboration layer realizes bidirectional communication through 5G private network.

[0019] The technical effects of the above embodiment include: through the deep coupling of edge computing nodes and numerical control system, the system realizes ≤2ms data fusion delay, which is reduced by more than 80% compared with traditional cloud scheme. Dynamic weight distribution algorithm can improve the fusion accuracy of sensor data, and adaptive sampling rate adjustment mechanism can reduce the amount of invalid data transmission.

[0020] The technical effects of the above embodiment include: Through the deep coupling of edge computing nodes and numerical control system, the system realizes ≤2ms data fusion delay, which is reduced by more than 80% compared with traditional cloud scheme. Dynamic weight distribution algorithm makes the fusion accuracy of sensor data improve by 35%, and adaptive sampling rate adjustment mechanism reduces the amount of invalid data transmission by 40%. The test shows that under the condition of variable depth milling, the recognition accuracy of the system to the main shaft bearing fault is 98.7%, and the false positive rate is less than 0.3%, which is significantly better than the existing technology.

[0021] Based on this, the edge computing node is coupled with the numerical control unit of the numerical control system through a PCIe Gen3x4 interface, and the data interaction period is set to 1 ms; the edge node is configured with a dual-redundant power module, and the switching time from the main power supply to the standby power supply is ≤15 ms when the main power supply fails; the edge computing node is internally provided with a hardware security module, which supports the AES-256 encryption algorithm to encrypt the transmission data in real time; and the node shell is made of an aluminum-based composite material.

[0022] It is worth mentioning that: the edge computing node is coupled with the numerical control unit (NCU) of the numerical control system through a PCIe Gen3x4 interface, and the data interaction period is set to 1 ms. A dual-redundant power module is configured, and the switching time from the main power supply to the standby power supply is ≤15 ms when the main power supply fails. A hardware security module (HSM) is internally provided, which supports the AES-256 encryption algorithm to encrypt the transmission data in real time. The node shell is made of an aluminum-based composite material with a thermal conductivity ≥200 W / (m·K), and the surface is anodized to a thickness ≥15 μm, meeting the IP67 protection level requirements.

[0023] The technical effects of the above embodiments include: The dual-redundant power module can improve system availability, the hardware security module can realize end-to-end encryption of data transmission, and the key rotation period is reduced. The aluminum-based composite material shell enables the node to operate stably in an environment of -20℃ to +70℃, and the IP67 protection level resists dust and short-term immersion.

[0024] The traditional technical solution has the following technical problems: the vibration sensor installation process is not standardized, resulting in signal distortion, the traditional cable design lacks sufficient anti-interference ability, the charge amplifier lacks gain adjustment function, and the base damping design is missing, causing high-frequency resonance.

[0025] Based on this, the three-way accelerometer of the vibration sensor array is fixed to the main shaft bearing seat through an M5 stainless steel threaded joint; the sensor signal line adopts a twisted shielded cable, the shielding layer ground resistance is ≤0.1Ω, and the cable bending radius is ≥10 times the cable outer diameter; each accelerometer is configured with an independent charge amplifier module; a 0.5mm thick silica gel damping layer is arranged between the mounting base and the machine tool body to isolate high-frequency vibration interference.

[0026] It is worth mentioning that: the three-way accelerometer of the vibration sensor array in this embodiment is fixed to the main shaft bearing seat through an M5 stainless steel threaded joint, and the mounting surface flatness is ≤0.01mm. The sensor signal line adopts a twisted shielded cable, the shielding layer ground resistance is ≤0.1Ω, and the cable bending radius is ≥10 times the cable outer diameter. Each accelerometer is configured with an independent charge amplifier module, the gain adjustment range is 1-1000 times, and the noise density is ≤0.5μV / √Hz. A 0.5mm thick silica gel damping layer is arranged between the mounting base and the machine tool body to isolate high-frequency vibration interference.

[0027] The technical effects of the above embodiment include that the M5 threaded joint and the 0.01 mm flatness mounting surface control the sensor inherent frequency error within ±1 Hz. The twisted shielded cable improves the common mode rejection ratio to more than 80 dB, and the independent charge amplifier module realizes dynamic adjustment of signal gain. The silica gel damping layer suppresses high-frequency vibration above 1 kHz, and the spindle bearing fault feature frequency extraction accuracy is improved by 50%.

[0028] The traditional technical solution has the following technical problems: the temperature sensor embedding position is unreasonable, leading to temperature measurement lag; the traditional compensation wire error is large; the lack of multiplexing switch causes signal crosstalk; and the shell material lacks corrosion resistance. Based on this, the three PT100 sensors of the temperature sensor group are embedded in the stator winding, the rotor shaft end and the front bearing outer ring of the main shaft motor respectively; the sensor lead adopts four-core compensation wire, and the cold end compensation is realized by a digital temperature sensor; each temperature sensor is configured with an independent multiplexing switch, and the sampling period can be programmed; the sensor shell is made of 316L stainless steel material.

[0029] It is worth mentioning that: the three PT100 sensors of the temperature sensor group are embedded in the stator winding, the rotor shaft end and the front bearing outer ring of the main shaft motor respectively. The sensor lead adopts four-core compensation wire, and the length error is ≤0.5%. The cold end compensation is realized by a digital temperature sensor. Each temperature sensor is configured with an independent multiplexing switch, and the sampling period can be programmed. The minimum sampling interval is 10 ms. The sensor shell is made of 316L stainless steel material, with a protection level of IP68 and a working temperature range of -50℃ to +250℃.

[0030] The technical effects of the above embodiment include: the three-point temperature measurement layout of the stator winding, the rotor shaft end and the bearing outer ring realizes three-dimensional reconstruction of the temperature field. The four-core compensation wire and the digital cold end compensation improve the temperature measurement accuracy to ±0.1℃, the independent multiplexing switch eliminates crosstalk between channels, and the sampling interval is shortened to 10 ms. The 316L stainless steel shell is not corroded after 480 hours of salt spray test, and the IP68 protection level resists high-pressure water washing.

[0031] The traditional technical solution has the following technical problems: the traditional entropy weight method does not consider the influence of sensor working environment temperature, the weight distribution lacks dynamic adaptability, and the fusion result is distorted under high temperature or low temperature working conditions.

[0032] Based on this, the dynamic weight distribution algorithm adopts an improved entropy weight method, and the weight coefficient calculation formula is: ; In the formula: is the dynamic weight coefficient of the jth sensor; is the information entropy, wherein , k is the normalization constant; is the preset reference temperature threshold, is the real-time temperature value of the jth sensor; n is the total number of sensors, and m is the sampling window length.

[0033] The formula consists of two core product terms: Normalized entropy weight: ; : represents the information entropy of the j-th sensor, where is the data probability distribution of the j-th sensor at the i-th sampling point, and k is the Boltzmann constant normalization term.

[0034] Information entropy The physical meaning of is to measure the uncertainty of sensor data: when When it approaches 0, it means that the data concentration is high and the weight should be increased; when When it approaches 1, the data is evenly distributed and the weight should be reduced.

[0035] Normalization ensures that .

[0036] Temperature compensation term: ; To preset the reference temperature threshold, usually the rated operating temperature of the spindle motor is used. is the real-time temperature value of the jth sensor.

[0037] This item introduces a temperature feedback mechanism: when the temperature of the sensor's working environment rises and causes the electronic components to drift, the influence of thermal noise is suppressed by reversely adjusting the weight. For example, when When the denominator increases, the weight Attenuation is used to avoid the interference of the sensor signal-to-noise ratio drop on the fusion result under high temperature conditions. During the CNC lathe processing, the characteristic frequency of the spindle bearing fault will change with the speed. change( , where d is the ball diameter and D is the pitch circle diameter. The traditional entropy weight method does not consider the temperature effect. In the scenario where continuous cutting causes the bearing temperature to rise, the entropy value of the vibration sensor data is Thermal expansion can produce spurious fluctuations. This formula dynamically adjusts the weights through temperature compensation terms, ensuring that the accuracy of fault characteristic frequency extraction remains within 2% even at a temperature rise of 30°C, a 60% improvement over traditional methods.

[0038] The technical effects of the above embodiment include: after the temperature compensation term is introduced, the weight coefficient can shorten the response time to the change of ambient temperature.

[0039] For example, traditional technical solutions have the following technical problems: Fixed sampling rate leads to data redundancy or loss of critical information, and traditional adaptive algorithms have slow response speed, which cannot meet the processing requirements under variable working conditions. The adaptive sampling rate adjustment mechanism is realized by the following formula: ; In the formula: is the adjusted sampling frequency; is the basic sampling rate; is the standard deviation of the current vibration signal, is the preset maximum allowed standard deviation; α is the adjustment coefficient, and β is the exponential factor. The formula realizes the dynamic matching of the sampling rate and the signal complexity through nonlinear mapping: Basic sampling rate term: , usually set to 1kHz, meets the minimum requirement of ISO 10816-3 vibration standard for machine tool spindle monitoring.

[0040] Dynamic adjustment term: ; : The standard deviation of the current vibration signal reflects the signal fluctuation intensity; : The preset maximum allowed standard deviation is determined by the dynamic characteristics of the machine tool, such as the spindle of a certain type of CNC lathe ; : Adjustment coefficient (0.8), controls the sampling rate adjustment amplitude; : Exponential factor (2.5), determines the steepness of the adjustment curve.

[0041] In the rough machining stage, such as milling depth , the standard deviation of the vibration signal may reach , at this time , the adjustment term is , and the sampling rate is increased to 1.46kHz. In the finishing stage (R ), the adjustment term tends to 1, and the sampling rate falls back to the basic value.

[0042] The technical effects of the above embodiment include: The above scheme can realize the response time ≤10ms of the sampling rate automatically increasing from 1kHz to 5kHz when the milling feed rate suddenly changes from 0.1mm / r to 0.5mm / r.

[0043] The traditional technical scheme has the following technical problems: limited edge computing resources lead to insufficient precision of complex fault diagnosis, high cloud diagnosis delay, and lack of dynamic adjustment ability of collaborative mechanism. Based on this, an edge-cloud collaborative working mode is also included, which adopts the following data fusion strategy: ; In the formula: For the final decision result; For the edge node preliminary diagnosis conclusion; For the cloud depth analysis result; λ is the confidence coefficient.

[0044] The formula realizes the dynamic matching of the sampling rate and the signal complexity through nonlinear mapping: the basic sampling rate term: Usually set to 1kHz, meet the minimum requirements of ISO 10816-3 vibration standard for machine tool spindle monitoring.

[0045] Dynamic adjustment term: ; : Current vibration signal standard deviation, reflecting the signal fluctuation intensity.

[0046] : The preset maximum allowable standard deviation is determined by the dynamic characteristics of the machine tool, such as the spindle of a certain type of CNC lathe ; : Adjustment coefficient 0.8, control sampling rate adjustment amplitude; : Exponential factor 2.5, determine the steepness of the adjustment curve.

[0047] In the rough machining stage, such as milling depth , the vibration signal standard deviation may reach , at this time , the adjustment term is , the sampling rate is raised to 1.46kHz. In the finishing stage , the adjustment term tends to 1, and the sampling rate falls back to the basic value.

[0048] The technical effects of the above embodiment include: Can improve the accuracy of edge node preliminary diagnosis, reduce the data transmission volume of collaborative mechanism, and reduce the load of the cloud.

[0049] The traditional technical solution has the following technical problems: the traditional SVM model parameter optimization is easy to fall into local optimum, the high model complexity causes large edge node reasoning delay, and the sample imbalance affects the minority class fault recognition rate. Based on this, when the support vector machine model adopts the improved grey wolf optimization algorithm for parameter optimization, the wolf population size is set to 30-50 individuals, and the maximum iteration number is 80-120 times; the fitness function is designed as the weighted sum of classification accuracy and model complexity, the weight coefficient is dynamically adjusted according to the cross-validation result, the population position is initialized by introducing chaos mapping to avoid falling into local optimum in the optimization process; the model training data set contains normal state, early fault and serious fault three types of samples, and the number ratio of each type of sample is kept at 3:1:1.

[0050] The technical effects of the above embodiments include: The chaos mapping initialization can improve the global search capability of the optimization algorithm and shorten the model inference delay. In the imbalanced sample data set, the minority class fault recognition rate is improved.

[0051] The traditional technical solution has the following technical problems: the traditional industrial Ethernet has single point failure risk, insufficient priority queue, low clock synchronization accuracy, slow fault recovery, etc.

[0052] Therefore, the time-sensitive network configuration includes dual redundant switches, the switching time of the master and standby switches is ≤500ns; 8 priority queues are allocated for sensor data streams, vibration data, temperature data and current data correspond to Class 3, Class 4 and Class 5 priority respectively; the gate list configuration is synchronized with the interpolation period of the numerical control system; the network fault recovery mechanism includes link aggregation control protocol and rapid spanning tree protocol.

[0053] It is worth mentioning that: the time-sensitive network configuration includes dual redundant switches, the switching time of the master and standby switches is ≤500ns. 8 priority queues are allocated for sensor data streams, vibration data, temperature data and current data correspond to Class 3, Class 4 and Class 5 priority respectively. The gate list configuration is synchronized with the interpolation period of the numerical control system, and the synchronization error is ≤1μs. The network fault recovery mechanism includes link aggregation control protocol (LACP) and rapid spanning tree protocol (RSTP), and the convergence time is ≤100ms.

[0054] The technical effects of the above embodiments include: dual redundant switches can improve network availability, and 8-level priority queues can realize fine QoS management. In the scenario where the interpolation period of the numerical control system is 4ms, the synchronization error of the gate list is ≤0.5μs. The combination of LACP+RSTP shortens the network convergence time from 2 seconds of traditional STP to 80ms.

[0055] The traditional technical solution has the following technical problems: the traditional Wi-Fi or 4G network has coverage blind area, insufficient bandwidth, low transmission reliability, poor service isolation, etc.

[0056] Therefore, the 5G private network uses 2.6GHz frequency band, and deploys distributed micro base stations to realize full coverage of the factory area; the user plane function is sunk to the edge machine room of the factory, and is connected with the core network through N3 interface; A dual-transmission and selective-reception transmission mechanism is configured, important control instructions are sent through two independent paths, and the receiving end selects the first arriving data packet; Network slicing technology divides three independent logical channels, which respectively bear sensor data, control instructions and video monitoring services.

[0057] It is worth mentioning that: The 5G private network in the above embodiment uses the 2.6 GHz frequency band and deploys distributed micro base stations to achieve full coverage of the factory area, with base station spacing ≤ 200 meters. The user plane function (UPF) is deployed in the edge equipment room of the factory area and connected to the core network via the N3 interface, with a transmission bandwidth of ≥ 1 Gbps. A dual-transmit selective receive (FRER) transmission mechanism is configured, and important control instructions are sent via two independent paths. The receiving end selects the first arriving data packet. Network slicing technology divides the network into three independent logical channels, respectively carrying sensor data, control instructions, and video surveillance services.

[0058] The technical benefits of the above-mentioned embodiments include: distributed micro base stations achieve 100% 5G coverage in the factory area, with an average signal strength of ≥-65dBm. UPF sinking reduces data transmission latency from 50ms in traditional 4G to 8ms, and a dual-transmit selective reception mechanism improves control command transmission reliability to 99.999%. Network slicing technology achieves inter-service isolation of ≥50dB, and in bursty video surveillance traffic scenarios, sensor data transmission bandwidth fluctuations are ≤5%.

[0059] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A multi-sensor fusion monitoring system for CNC lathes based on edge computing, characterized in that: include: Edge computing layer: The edge computing node deployed in the CNC system control cabinet is equipped with a hardware platform including a quad-core ARM Cortex-A72 processor, 16GB DDR4 memory and NVMe solid-state drive, running a real-time operating system; Sensor network layer: A multimodal sensing unit consisting of a vibration sensor array, a temperature sensor group, a current sensor, and an acoustic emission sensor. The vibration sensor uses a three-axis accelerometer arranged on the surface of the spindle bearing seat, and the temperature sensor uses a PT100 platinum resistor attached to the spindle motor housing. The sensor network layer also includes an adaptive sampling rate adjustment mechanism for dynamically adjusting the sampling frequency based on signal characteristics. Data fusion layer: Implements a dynamic weight allocation algorithm to automatically adjust the fusion weight coefficient of each sensor data according to the current processing stage; Abnormal decision layer: Establish a lightweight fault classification model based on support vector machine, with the model input dimension not exceeding 12 dimensions; Communication layer: Use time-sensitive network protocol to achieve deterministic transmission of sensor data and control instructions; Cloud collaboration layer: Establishes two-way communication with edge nodes through the 5G private network, triggering the cloud-based deep diagnosis process only when a potential fault is detected.

2. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The edge computing node is coupled to the CNC system's CNC unit at the hardware level via a PCIe Gen3×4 interface, with a data interaction cycle set to 1ms. The edge node is equipped with dual redundant power supply modules, and the switching time to the backup power supply is ≤15ms when the main power supply fails. The edge computing node has a built-in hardware security module that supports the AES-256 encryption algorithm for real-time encryption of transmitted data. The node casing is made of aluminum-based composite material.

3. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The three-axis accelerometer of the vibration sensor array is fixed to the spindle bearing seat via an M5 stainless steel threaded joint. The sensor signal line uses a twisted-pair shielded cable with a shielding layer grounding resistance of ≤0.1Ω and a cable bending radius of ≥10 times the cable outer diameter. Each accelerometer is equipped with an independent charge amplifier module. A 0.5mm thick silicone damping layer is provided between the mounting base and the machine tool body to isolate high-frequency vibration interference.

4. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The three PT100 sensors of the temperature sensor group are respectively embedded in the stator winding of the spindle motor, the rotor shaft end and the outer ring of the front bearing; the sensor leads use four-core compensation wires, and cold-end compensation is achieved using a digital temperature sensor; each temperature sensor is equipped with an independent multiplexing switch, and the sampling period is programmable; the sensor housing is made of 316L stainless steel.

5. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The dynamic weight allocation algorithm adopts the improved entropy weight method, and the weight coefficient calculation formula is: ； Where: is the dynamic weight coefficient of the jth sensor; is the information entropy, where , k is the normalization constant; is the preset reference temperature threshold, is the real-time temperature value of the jth sensor; n is the total number of sensors, and m is the sampling window length.

6. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The adaptive sampling rate adjustment mechanism is implemented by the following formula: ； Where: is the adjusted sampling frequency; is the base sampling rate; is the current vibration signal standard deviation, is the preset maximum allowable standard deviation; α is the adjustment coefficient, and β is the exponential factor.

7. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: It also includes an edge-cloud collaborative working mode, which adopts the following data fusion strategy: ； Where: For the final decision result; Provides preliminary diagnostic conclusions for edge nodes; is the cloud-based deep analysis result; λ is the confidence coefficient.

8. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: When the support vector machine model uses the improved gray wolf optimization algorithm for parameter optimization, the wolf pack size is set to 30-50 individuals, and the maximum number of iterations is 80-120 times; the fitness function is designed to be the weighted sum of classification accuracy and model complexity, and the weight coefficient is dynamically adjusted according to the cross-validation results to introduce a chaotic map to initialize the population position to avoid the optimization process falling into a local optimum; the model training data set contains three types of samples: normal state, early fault, and severe fault, and the ratio of the number of samples in each category is maintained at 3:1:

1.

9. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1 is characterized in that: The time-sensitive network configuration includes dual redundant switches, with the switching time between the primary and standby switches ≤ 500ns; eight priority queues are assigned to sensor data streams, with vibration data, temperature data, and current data corresponding to Class 3, Class 4, and Class 5 priorities, respectively; the gate list configuration is synchronized with the CNC system interpolation cycle; and the network fault recovery mechanism includes the Link Aggregation Control Protocol and the Rapid Spanning Tree Protocol.

10. The multi-sensor fusion monitoring system for CNC lathes based on edge computing according to claim 1, characterized in that: The 5G private network uses the 2.6GHz frequency band and deploys distributed micro base stations to achieve full coverage of the factory area; User plane functions are moved down to the edge equipment room in the factory area and connected to the core network via the N3 interface. Configure a dual-send selective reception transmission mechanism, where important control instructions are sent through two independent paths, and the receiver selects the data packet that arrives first; Network slicing technology divides the network into three independent logical channels, which respectively carry sensor data, control instructions and video surveillance services.

Images