Shield tunneling machine cutterhead motor fault monitoring and early warning device and early warning method thereof

Abstract

The invention discloses a fault monitoring and early warning device for a cutter head motor of a shield tunneling machine. Comprising temperature sensors installed on a stator winding, a rotor winding and a bearing of the cutter head motor, a voltage and current sensor installed on the cutter head motor, a vibration sensor installed on the bearing of the cutter head motor, and a smoke sensor installed at a ventilation opening or the top of a motor shell. The noise sensor is arranged on the surface of the motor shell close to a bearing or a rotor part, the temperature sensor, the voltage and current sensor, the vibration sensor, the smoke sensor and the noise sensor are all connected with a microcontroller unit MCU through an SPI bus, and the microcontroller unit MCU is also connected with an alarm device. According to the invention, the problem that fault discovery lags due to the fact that real-time and continuous fault early warning cannot be realized through manual regular inspection in the prior art is solved. The invention further discloses a fault early warning method for the cutter head motor of the shield tunneling machine.

Description

Technical Field

[0001] The present invention belongs to the technical field of shield machine fault warning devices, and relates to a fault monitoring and warning device for the cutter head motor of a shield machine. The present invention also relates to a method for fault warning using the above-mentioned fault monitoring and warning device for the cutter head motor of a shield machine. Background Art

[0002] When a shield machine operates under certain working conditions, the motor is prone to various problems due to environmental conditions and load changes. During subway tunnel construction, the alternation of soft soil and sand layers and the high-humidity environment are likely to cause deterioration of the motor insulation performance, bearing wear, and current overload. In water conservancy tunnel projects, high water pressure, high humidity, and hard rock formations may lead to blockage of the cooling system, shaft current corrosion, and rotor dynamic balance damage. In mine roadway tunneling, the working conditions of high dust, high vibration, and the alternation of hard rock and fault zones are likely to cause harmonic overheating, dust intrusion, and mechanical impact damage. At present, the fault monitoring of the cutter head motor of a shield machine mainly relies on the data collection of a single sensor, such as a temperature sensor or a vibration sensor. However, a single sensor cannot comprehensively reflect the operating state of the motor. Especially under complex geological conditions, the monitoring of a single data source is prone to misjudgment or missed judgment. In addition, most of the existing monitoring systems rely on manual regular inspections and cannot achieve real-time and continuous fault warning, resulting in a lag in fault discovery and increasing the risk of equipment damage and construction delays. Summary of the Invention

[0003] The purpose of the present invention is to provide a fault monitoring and warning device for the cutter head motor of a shield machine, which solves the problem in the prior art that real-time and continuous fault warning cannot be achieved through manual regular inspections, resulting in a lag in fault discovery.

[0004] The present invention also discloses a method for fault warning using the above-mentioned fault monitoring and warning device for the cutter head motor of a shield machine.

[0005] The technical solution adopted by the present invention is that a fault monitoring and warning device for the cutter head motor of a shield machine includes temperature sensors installed on the stator winding, rotor winding, and bearings of the cutter head motor, voltage and current sensors installed on the cutter head motor, vibration sensors installed on the bearings of the cutter head motor, smoke sensors installed at the ventilation openings or the top of the motor housing, and noise sensors installed on the surface of the motor housing near the bearings or the rotor. The temperature sensors, voltage and current sensors, vibration sensors, smoke sensors, and noise sensors are all connected to a microcontroller unit MCU through an SPI bus, and the microcontroller unit MCU is also connected to an alarm device.

[0006] Preferably, the microcontroller unit MCU is equipped with a data processing module, a data fusion module, and a trained convolutional neural network.

[0007] Preferably, the microcontroller unit MCU is also connected to a PWM speed regulation module, and the PWM speed regulation module is connected to the drive circuit of the cutter head motor of the shield machine.

[0008] The second technical solution adopted by the present invention is a fault monitoring and early warning method for the cutter head motor of a shield machine. The above-mentioned fault monitoring and early warning device for the cutter head motor of a shield machine is used, specifically: the temperature sensor, voltage and current sensor, vibration sensor, smoke sensor, and noise sensor collect the temperature, voltage, current, vibration, smoke concentration, and noise data corresponding to the operation data of the cutter head motor of the shield machine in real time, and transmit the data to the microcontroller unit MCU in real time. After being processed by the data processing module in the microcontroller unit MCU, the microcontroller unit MCU is set with a temperature threshold, current threshold, voltage threshold, smoke or harmful gas concentration threshold, vibration amplitude and frequency threshold, and noise threshold. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU controls the alarm device to give an alarm. If none of them exceed the corresponding threshold, the corresponding data is subjected to feature fusion in the data fusion module, and the fused features are input into the convolutional neural network for fault diagnosis. If no fault is detected, the motor continues to operate normally. If the diagnosis result is a fault, corresponding intervention operations are performed according to the fault type.

[0009] Preferably, the temperature sensor, voltage and current sensor, vibration sensor, and noise sensor process the collected data in the data processing module. The processed temperature, voltage, current, vibration, and noise data are compared with the corresponding thresholds. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU controls the alarm device to give an alarm; the smoke sensor directly compares the collected data with the smoke or harmful gas concentration threshold in the microcontroller unit MCU. If the collected value is greater than the smoke or harmful gas concentration threshold, the microcontroller unit MCU controls the alarm device to give an alarm.

[0010] Preferably, the data processing module performs frequency domain analysis on the vibration signals and noise signals collected by the vibration sensor and noise sensor using Fourier transform to extract spectral features; then, the processed vibration signals and noise signals, as well as the temperature data and current and voltage data collected by the temperature sensor and voltage and current sensor, are processed to remove outliers, and interpolation or fitting methods are used to supplement missing data to ensure data continuity;

[0011] The microcontroller unit MCU performs data fusion on the data of the temperature sensor, voltage and current sensor, vibration sensor, and noise sensor processed by the data processing module in the data fusion module to obtain the fused features, and then inputs the fused features into the convolutional neural network for fault diagnosis. If the diagnosis result is a fault, corresponding intervention operations are performed according to the fault type, that is, corresponding control signals are sent to the PWM speed regulation module through the microcontroller unit. The PWM speed regulation module is connected to the shield machine cutter head motor drive circuit to dynamically adjust the motor operating state.

[0012] Preferably, the fused feature F fused Specifically:

[0013]

[0014] where x i is the data of the i-th sensor, i = 1 - 7, respectively representing the temperature data collected by three temperature sensors, the voltage and current data collected by the voltage and current sensor, the vibration signal of the vibration sensor, and the noise signal collected by the noise sensor. w j is the weight of the corresponding sensor data under the j-th type of fault, and K is the total number of fault types.

[0015] Preferably, the convolutional neural network is trained before use, and the training process is implemented according to the following steps:

[0016] Step 1, obtain historical fault data of different fault types. Specifically: obtain the data collected by the corresponding temperature sensor, voltage and current sensor, vibration sensor, and noise sensor when the corresponding fault occurs, that is, each group of data includes the temperature data collected by three temperature sensors, the voltage and current data collected by the voltage and current sensor, the vibration signal collected by the vibration sensor, and the noise signal collected by the noise sensor. Each group of data is marked with the corresponding label, that is, the corresponding fault type;

[0017] Step 2, perform outlier removal processing on the data collected in Step 1, and use interpolation or fitting methods to supplement missing data to ensure data continuity; and perform frequency domain analysis on the noise signals and vibration signals collected by the vibration sensor and noise sensor to extract spectral features, obtain the final sample data, and divide the sample data into a training set and a test set;

[0018] Step 3, perform data fusion on each group of data in the training set using the dynamic weight allocation strategy based on the entropy value method to obtain the fused feature F fused ;

[0019] Step 4, set training parameters, and input the fused feature F fused into the convolutional neural network for deep learning to identify the fault type:

[0020] Fault type = softmax(W fc ·(W conv *F fused ))

[0021] where W conv is the convolution kernel, * represents the convolution operation, W fc is the fully connected weight, and softmax is the activation function, which is used to calculate the probability distribution of fault categories;

[0022] Thus, a convolutional neural network model capable of identifying fault types is obtained.

[0023] Preferably, the fused feature F in step 3 fused is specifically:

[0024]

[0025] where x i is the data of the i-th sensor, i = 1 - 7, which respectively represent the temperature data collected by three temperature sensors, the voltage and current data collected by voltage and current sensors, the vibration signal of the vibration sensor, and the noise signal collected by the noise sensor, w j is the weight of the corresponding sensor data under the j-th fault type, and K is the total number of fault types;

[0026] where w j is calculated according to the following formula:

[0027]

[0028] where e j is the entropy value of the corresponding sensor data under the j-th fault type;

[0029]

[0030] where the constant n is the number of data in each group, n = 7;

[0031] p ij represents the information contribution ratio of the data of the i-th sensor in the j-th fault type, and is calculated according to the following formula:

[0032]

[0033] Preferably, when dividing the training set and the test set in step 2, the division is performed according to the following method:

[0034] Let the total sample set D contain K types of fault data, and the number of samples of each type of fault is N k , then the proportion of the training set is α, then:

[0035]

[0036] Among them, is the training set, is the test set;

[0037] Ensure by random sampling without replacement:

[0038]

[0039] The beneficial effects of the present invention are:

[0040] The present invention detects the temperature, noise, voltage, current and bearing vibration state of the motor through sensors, can realize real-time and continuous data acquisition, overcome the disadvantages of the time interval of manual detection, capture the subtle changes of data in time, fuse the collected data based on the multi-modal data fusion algorithm, and then accurately judge the running state of the motor through the trained convolutional neural network, greatly improving the efficiency and accuracy of fault diagnosis. Once abnormal data is detected, the early warning mechanism can be automatically triggered immediately to notify relevant personnel to take measures to avoid the further deterioration of the fault, effectively reducing the downtime and maintenance costs, and ensuring the stable operation of the motor and the continuity of production activities. Description of the Drawings

[0041] Figure 1 is a schematic structural diagram of the fault monitoring and early warning device for the cutter head motor of the shield machine of the present invention;

[0042] Figure 2 is a flowchart of the fault monitoring and early warning method for the cutter head motor of the shield machine of the present invention.

[0043] In the figure: 1. Temperature sensor, 2. Voltage and current sensor, 3. Vibration sensor, 4. Smoke sensor, 5. Noise sensor, 6. Microcontroller unit MCU, 7. Alarm device. Detailed Embodiments

[0044] The following is a detailed description in conjunction with the specific embodiments.

[0045] Embodiment 1

[0046] The fault monitoring and early warning device for the cutter head motor of the shield machine of the present invention has a structure as Figure 1As shown, it includes a temperature sensor 1 installed on the stator winding, rotor winding, and bearings of the cutter head motor, a voltage and current sensor 2 installed on the cutter head motor, a vibration sensor 3 installed on the bearings of the cutter head motor, a smoke sensor 4 installed at the ventilation opening or the top of the motor housing, and a noise sensor 5 installed on the surface of the motor housing near the bearings or the rotor. The temperature sensor 1, voltage and current sensor 2, vibration sensor 3, smoke sensor 4, and noise sensor 5 are all commonly connected to a microcontroller unit MCU6 through the SPI bus, and the microcontroller unit MCU6 is also connected to an alarm device 7.

[0047] The microcontroller unit MCU6 is equipped with a data processing module, a data fusion module, and a trained convolutional neural network.

[0048] The microcontroller unit MCU6 is also connected to a PWM speed regulation module, and the PWM speed regulation module is connected to the drive circuit of the shield machine cutter head motor.

[0049] Embodiment 2

[0050] The fault monitoring and early warning method for the cutter head motor of the shield machine of the present invention has a process as Figure 2 shown. The fault monitoring and early warning device for the cutter head motor of the shield machine in Embodiment 1 is adopted. Specifically: The temperature sensor 1, voltage and current sensor 2, vibration sensor 3, smoke sensor 4, and noise sensor 5 collect the temperature, voltage, current, vibration, smoke concentration, and noise data corresponding to the operation data of the shield machine cutter head motor in real time, and transmit the data to the microcontroller unit MCU6 in real time. It is processed by the data processing module in the microcontroller unit MCU6. The microcontroller unit MCU6 is set with a temperature threshold, a current threshold, a voltage threshold, a smoke or harmful gas concentration threshold, a vibration amplitude and frequency threshold, and a noise threshold. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU6 controls the alarm device 7 to give an alarm. If none of them exceed the corresponding threshold, the corresponding data is subjected to feature fusion in the data fusion module, and the fused features are input into the convolutional neural network for fault diagnosis. If no fault is detected, the motor continues to operate normally. If the diagnosis result is that a fault has occurred, corresponding intervention operations are performed according to the type of fault.

[0051] Embodiment 3

[0052] On the basis of Embodiment 2, the vibration amplitude and frequency threshold are divided into a vibration amplitude threshold and a frequency threshold. When either the vibration amplitude or the frequency of the vibration signal collected by the vibration sensor 3 exceeds the corresponding threshold, an alarm is given.

[0053] Embodiment 4

[0054] Based on Embodiment 3, the temperature sensor 1, voltage-current sensor 2, vibration sensor 3, and noise sensor 5 process the collected data in the data processing module. The processed temperature, voltage, current, vibration, and noise data are compared with the corresponding thresholds. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU6 controls the alarm device 7 to give an alarm; the smoke sensor 4 directly compares the collected data with the smoke or harmful gas concentration threshold in the microcontroller unit MCU6. If the collected value is greater than the smoke or harmful gas concentration threshold, the microcontroller unit MCU6 controls the alarm device 7 to give an alarm.

[0055] The data processing module performs frequency-domain analysis on the vibration signals and noise signals collected by the vibration sensor 3 and noise sensor 5 using Fourier transform to extract spectral features; then, the processed vibration signals, noise signals, temperature data collected by the temperature sensor 1, and current and voltage data are processed to remove outliers, and interpolation or fitting methods are used to supplement missing data to ensure data continuity;

[0056] Among them, performing frequency-domain analysis on the vibration signals and noise signals using Fourier transform to extract spectral features specifically means:

[0057] If the vibration signals and noise signals are time-domain signals x(t), then perform Fourier transform on the time-domain signal x(t) to obtain the frequency-domain representation X(f):

[0058]

[0059] Where: X(f) is the frequency-domain signal, x(t) is the time-domain signal, and f is the frequency.

[0060] Extract spectral features from the Fourier transform result X(f), which are divided into the amplitude spectrum and the phase spectrum:

[0061] A(f) = |X(f)|

[0062] Where: A(f) is the amplitude at frequency f, and |X(f)| is the modulus of X(f).

[0063] φ(f) = arg(X(f))

[0064] Where: φ(f) is the phase at frequency f, and arg(X(f)) is the phase angle of X(f).

[0065] The microcontroller unit MCU6 fuses the data of the temperature sensor 1, voltage and current sensor 2, vibration sensor 3, and noise sensor 5 processed by the data processing module within the data fusion module to obtain the fused features, and then inputs the fused features into the convolutional neural network for fault diagnosis. If the diagnosis result indicates a fault, corresponding intervention operations are performed according to the fault type, that is, corresponding control signals are sent to the PWM speed regulation module through the microcontroller unit. The PWM speed regulation module is connected to the shield machine cutter head motor drive circuit to dynamically adjust the motor operating state.

[0066] Embodiment 5

[0067] Based on Embodiment 4, the fused feature F fused Specifically:

[0068]

[0069] where x i is the data of the i-th sensor, i = 1 - 7, respectively representing the temperature data collected by the three temperature sensors 1, the voltage and current data collected by the voltage and current sensor 2, the vibration signal of the vibration sensor 3, and the noise signal collected by the noise sensor 5. w j is the weight of the corresponding sensor data under the j-th type of fault, and K is the total number of fault types.

[0070] Embodiment 6

[0071] Based on Embodiment 5, the convolutional neural network is trained before use, and the training process is implemented according to the following steps:

[0072] Step 1, obtain historical fault data of different fault types. Specifically: obtain the data collected by the corresponding temperature sensor 1, voltage and current sensor 2, vibration sensor 3, and noise sensor 5 when the corresponding fault occurs, that is, each set of data includes the temperature data collected by the three temperature sensors 1, the voltage and current data collected by the voltage and current sensor 2, the vibration signal collected by the vibration sensor 3, and the noise signal collected by the noise sensor 5. Each set of data is marked with the corresponding label, that is, the corresponding fault type;

[0073] Step 2, perform outlier removal processing on the data collected in Step 1, and use interpolation or fitting methods to supplement missing data to ensure data continuity; and perform frequency domain analysis on the noise signals and vibration signals collected by the vibration sensor 3 and noise sensor 5 to extract spectral features to obtain the final sample data. The sample data is divided into a training set and a test set, and the division is performed according to the following method:

[0074] Assume that the total sample set D contains K types of fault data, and the number of samples for each type of fault is N k, if the proportion of the training set is α, then:

[0075]

[0076] Among them, is the training set, is the test set;

[0077] Ensure by random sampling without replacement:

[0078]

[0079] Step 3, adopt a dynamic weight allocation strategy based on the entropy value method, and dynamically adjust the weights of parameters such as temperature, vibration, noise, voltage, and current according to the entropy values (information contribution degrees) of each parameter in the historical fault data; specifically, the entropy value reflects the information contribution degree of each parameter to fault diagnosis, and the smaller the entropy value, the greater the weight. For example, in hard rock formations, the entropy value of vibration data is small, and the weight is increased to 0.3; while in a high humidity environment, the entropy value of temperature data decreases significantly, and the weight is increased to 0.4 to more accurately reflect the actual working conditions. Through dynamic weight allocation, the system can adapt to the fault characteristics under different working conditions. For example, when the weights of vibration and noise are high, it may indicate bearing wear or rotor imbalance; when the weights of temperature and current are high, it may indicate winding overheating or overload. This dynamic weight adjustment mechanism significantly improves the accuracy and adaptability of fault diagnosis.

[0080] Adopt a dynamic weight allocation strategy based on the entropy value method to fuse each group of data in the training set to obtain the fused feature F fused , specifically:

[0081]

[0082] Among them, x i is the data of the i-th sensor, i = 1 - 7, respectively representing the temperature data collected by three temperature sensors 1, the voltage and current data collected by voltage and current sensor 2, the vibration signal of vibration sensor 3, and the noise signal collected by noise sensor 5, w j is the weight of the corresponding sensor data under the j-th fault type, and K is the total number of fault types;

[0083] Among them, w j is calculated according to the following formula:

[0084]

[0085] Among them, e j is the entropy value of the corresponding sensor data under the j-th fault type;

[0086]

[0087] Among them, the constant n is the number of data in each group of data, n = 7;

[0088] p ij represents the information contribution ratio of the data of the i-th sensor in the j-th type of fault, and is calculated according to the following formula:

[0089]

[0090] Step 4, set the training parameters, and input the fused feature F fused into the convolutional neural network for deep learning to identify the fault type:

[0091] Fault type = softmax(W fc ·(W conv *F fused ))

[0092] Among them, W conv is the convolutional kernel, * represents the convolutional operation, W fc is the fully connected weight, and softmax is the activation function, which is used to calculate the probability distribution of the fault category;

[0093] Thus, a convolutional neural network model capable of identifying the fault type is obtained.

[0094] Embodiment 7

[0095] On the basis of Embodiment 6, the temperature sensor is used to measure the temperatures of the stator winding, rotor winding and bearing of the cutter head motor of the shield machine in real time, and save the data in the microcontroller unit MCU. When the temperature reaches the set threshold (such as the stator winding temperature exceeds 130 °C or the bearing temperature exceeds 90 °C), the microcontroller unit MCU immediately controls the motor to stop running, starts the alarm device, and at the same time triggers the cooling system to reduce the temperature, avoiding serious faults such as insulation aging, bearing lubrication failure or winding burnout caused by overheating. In addition, the microcontroller unit MCU can also dynamically adjust the load state of the motor according to the temperature change trend, such as reducing the cutter head speed or the propulsion force, to ensure that the motor operates efficiently within the safe temperature range.

[0096] The voltage and current detection part plays a crucial role during the operation of the cutter head motor of the shield machine. Through the voltage and current, the load status and power supply quality of the motor can be judged in real time. For example, when the cutter head of the shield machine is tunneling in clay or hard rock strata, if it is detected that the current suddenly increases beyond the rated value (overcurrent phenomenon), or the voltage fluctuation exceeds the specified range (overvoltage situation), the system will respond quickly and trigger a protection mechanism (such as reducing the cutter head speed or emergency shutdown), thereby effectively preventing the motor from being damaged due to overcurrent or overvoltage. In addition, this part can also detect the current unbalance degree, warning of potential faults such as broken rotor bars or power supply phase loss, ensuring the stable and safe operation of the cutter head motor of the shield machine under complex geological conditions.

[0097] The vibration sensor is installed at the bearing part of the motor because the vibration of the bearing can intuitively and accurately reflect the overall operating state of the motor, especially under the complex working conditions of the shield machine (such as hard rock tunneling or soft soil propulsion). After installation, the vibration sensor plays a core role in accurately measuring the vibration amplitude, frequency characteristics and vibration direction of the motor in real time. Whether it is the slight vibration during normal operation or the abnormal vibration changes caused by bearing wear, rotor imbalance or mechanical alignment problems, it can be quickly captured and recorded. These vibration frequency data provide a key basis for evaluating and analyzing the operating state of the motor, helping the staff to detect potential problems in time and take targeted measures to ensure the stable and safe operation of the cutter head motor of the shield machine under complex geological conditions.

[0098] The smoke detection part is used to detect combustible gases, smoke and harmful gases such as carbon monoxide in the environment around the cutter head motor of the shield machine. During the construction of the shield machine, components such as the insulation material and winding inside the motor may catch fire due to overheating, overload or short circuit, generating smoke. This smoke will not only further damage the internal structure of the motor (such as carbonization of the insulation layer or burning of the winding), resulting in a decline in motor performance or even complete failure, but may also contain a large amount of harmful gases. Therefore, when the concentration of smoke or harmful gases collected by the smoke sensor reaches the set threshold, the microcontroller unit MCU will immediately control the motor to stop running, start an alarm device (such as audible and visual alarm or SMS notification), and link the ventilation system to discharge the harmful gases to maximize the protection of the lives of construction workers and the property safety of the shield machine equipment.

[0099] The present invention collects the operating data of the motor in real time through temperature sensors, voltage and current sensors, vibration sensors, smoke sensors and noise sensors, and performs data storage, processing and fault diagnosis. Among them, the fault diagnosis uses a convolutional neural network CNN based on a multi-modal data fusion algorithm, which can accurately identify fault types such as bearing wear, rotor imbalance, and insulation aging.

[0100] Through multimodal data fusion and convolutional neural network analysis, the present invention can achieve early fault warning of the cutter head motor of a shield machine, significantly reduce unplanned downtime, extend the service life of the equipment, and improve the safety and efficiency of tunnel construction at the same time.

Claims

1. Shield machine cutter head motor fault monitoring and early warning device, characterized in that, Including a temperature sensor (1) installed on the stator winding, rotor winding and bearings of the cutter head motor, a voltage and current sensor (2) installed on the cutter head motor, a vibration sensor (3) installed on the bearings of the cutter head motor, a smoke sensor (4) installed at the ventilation opening or the top of the motor housing, and a noise sensor (5) installed on the surface of the motor housing near the bearings or the rotor. The temperature sensor (1), voltage and current sensor (2), vibration sensor (3), smoke sensor (4), and noise sensor (5) are all commonly connected to a microcontroller unit MCU (6) through the SPI bus, and the microcontroller unit MCU (6) is also connected to an alarm device (7).

2. The shield machine cutter head motor fault monitoring and early warning device according to claim 1, characterized in that, A data processing module, a data fusion module and a trained convolutional neural network are installed in the microcontroller unit MCU (6).

3. The shield machine cutter head motor fault monitoring and early warning device according to claim 2, characterized in that, The microcontroller unit MCU (6) is also connected to a PWM speed regulation module, and the PWM speed regulation module is connected to the drive circuit of the shield machine cutter head motor.

4. Fault monitoring and early warning method for the cutter head motor of a shield machine, characterized in that, Using the shield machine cutter head motor fault monitoring and early warning device described in claim 3, specifically: the temperature sensor (1), voltage and current sensor (2), vibration sensor (3), smoke sensor (4), and noise sensor (5) collect the temperature, voltage, current, vibration, smoke concentration, and noise data corresponding to the operation data of the shield machine cutter head motor in real time, and transmit the data to the microcontroller unit MCU (6) in real time. The data is processed by the data processing module in the microcontroller unit MCU (6). The microcontroller unit MCU (6) is set with temperature thresholds, current thresholds, voltage thresholds, smoke or harmful gas concentration thresholds, vibration amplitude and frequency thresholds, and noise thresholds. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU (6) controls the alarm device (7) to give an alarm. If none of them exceed the corresponding thresholds, the corresponding data is subjected to feature fusion in the data fusion module, and the fused features are input into the convolutional neural network for fault diagnosis. If no fault is detected, the motor continues to operate normally. If the diagnosis result is that a fault has occurred, corresponding intervention operations are performed according to the fault type.

5. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 4, wherein, The temperature sensor (1), voltage and current sensor (2), vibration sensor (3), and noise sensor (5) process the collected data in the data processing module, and the processed temperature, voltage, current, vibration, and noise data are judged against the corresponding thresholds. When the value detected by the corresponding sensor is greater than or equal to the corresponding threshold, the microcontroller unit MCU (6) controls the alarm device (7) to give an alarm; the smoke sensor (4) directly compares the collected data with the smoke or harmful gas concentration threshold in the microcontroller unit MCU (6). If the collected value is greater than the smoke or harmful gas concentration threshold, the microcontroller unit MCU (6) controls the alarm device (7) to give an alarm.

6. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 5, characterized in that, The data processing module performs frequency-domain analysis on the vibration signals and noise signals collected by the vibration sensor (3) and the noise sensor (5) using Fourier transform to extract spectral features; then, it performs outlier removal on the processed vibration signals, noise signals, and the temperature data and current and voltage data collected by the temperature sensor (1) and the voltage and current sensor (2), and uses interpolation or fitting methods to supplement missing data to ensure data continuity. The microcontroller unit MCU (6) fuses the corresponding data of the temperature sensor (1), the voltage and current sensor (2), the vibration sensor (3), and the noise sensor (5) processed by the data processing module in the data fusion module to obtain the fused features, and then inputs the fused features into the convolutional neural network for fault diagnosis. If the diagnosis result indicates a fault, corresponding intervention operations are performed according to the fault type, that is, corresponding control signals are sent to the PWM speed regulation module through the microcontroller unit. The PWM speed regulation module is connected to the cutter head motor drive circuit of the shield machine to dynamically adjust the operating state of the motor.

7. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 6, wherein, The fused feature F fused Specifically: where x i is the data of the i-th sensor, where i = 1 - 7, respectively representing the temperature data collected by three temperature sensors (1), the voltage and current data collected by a voltage-current sensor (2), the vibration signal of a vibration sensor (3), and the noise signal collected by a noise sensor (5), w j is the weight of the corresponding sensor data under the j-th fault type, and K is the total number of fault types.

8. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 7, characterized in that, The convolutional neural network is trained before use, and the training process is implemented according to the following steps: Step 1, obtain historical fault data of different fault types. Specifically: obtain the data collected by the corresponding temperature sensor (1), voltage and current sensor (2), vibration sensor (3), and noise sensor (5) when the corresponding fault occurs. That is, each set of data includes the temperature data collected by three temperature sensors (1), the voltage and current data collected by the voltage and current sensor (2), the vibration signals collected by the vibration sensor (3), and the noise signals collected by the noise sensor (5). Each set of data is marked with the corresponding label, that is, the corresponding fault type. Step 2, perform outlier removal on the data collected in Step 1, and use interpolation or fitting methods to supplement missing data to ensure data continuity; and perform frequency-domain analysis on the noise signals and vibration signals collected by the vibration sensor (3) and the noise sensor (5) to extract spectral features, obtain the final sample data, and divide the sample data into a training set and a test set. Step 3: Apply the dynamic weight assignment strategy based on the entropy method to each group of data in the training set for data fusion, and obtain the fused feature F fused ; Step 4, set the training parameters and input the fused feature F fused into the convolutional neural network for deep learning to identify the fault type: Fault type = softmax(W fc ·(W conv *F fused )) Among them, W conv is the convolution kernel, * represents the convolution operation, and W fc is the fully connected weight, and softmax is the activation function, which is used to calculate the probability distribution of the fault categories; Thus, a convolutional neural network model capable of identifying fault types is obtained.

9. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 8, characterized in that The fused feature F in step 3 fused Specifically: where x i is the data of the i-th sensor, where i = 1 - 7, representing respectively the temperature data collected by three temperature sensors (1), the voltage and current data collected by a voltage and current sensor (2), the vibration signal of a vibration sensor (3), and the noise signal collected by a noise sensor (5), w j is the weight of the corresponding sensor data under the j-th fault type, and K is the total number of fault types; where w j is calculated according to the following formula: where, e j is the entropy value of the corresponding sensor data under the j-th type of fault; Among them, the constant n is the number of data in each group of data, n = 7; p ij represents the information contribution ratio of the data of the i-th sensor in the j-th type of fault, and is calculated according to the following formula:

10. The fault monitoring and early warning method for the cutter head motor of a shield machine according to claim 9, characterized in that When dividing the training set and the test set in Step 2, the following method is used for division: Suppose the total sample set D contains K types of fault data, and the number of samples for each type of fault is N k , and the proportion of the training set is α, then: Among them, is the training set, is the test set; Ensure by random sampling without replacement:

Images