Online gear lubrication state monitoring device and method based on deep learning
The online monitoring device uses deep learning to analyze electrical signals from helicopter gears, addressing the limitations of existing methods by providing real-time, accurate lubrication status assessment and reducing maintenance through self-powered operation.
Patent Information
- Application Number
- CN202410999422.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-12-03
AI Technical Summary
Existing methods for monitoring helicopter gear lubrication status, such as vibration, temperature, and oil analysis, are prone to errors, complex, and lack real-time capabilities, leading to high maintenance costs and potential safety hazards.
An online monitoring device using deep learning to collect electrical signals from gear surfaces, combined with a self-powered system and neural network analysis, enabling real-time lubrication status assessment and self-driving operation.
The device provides accurate, real-time monitoring of gear lubrication status, reducing maintenance costs and extending equipment life by detecting potential issues proactively.
Smart Images

Figure CN119063995A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to an online monitoring device and method for gear lubrication status based on deep learning. Background Art
[0002] Helicopter gears are key components in the helicopter transmission system, responsible for efficiently transmitting the engine power to the rotor, tail rotor and other auxiliary systems. It adjusts the speed and torque through the precise meshing of a series of gears to meet the needs of different flight and operating conditions. Helicopter gears must cope with extreme working environments, including high loads, high-speed rotation and severe vibrations, while also being lightweight, highly reliable and having a long life. Due to the extremely high precision and durability required in their design and manufacturing, any failure will result in serious safety hazards and high maintenance costs. Therefore, ensuring their trouble-free operation is crucial to flight safety, which makes advanced monitoring technology particularly urgent.
[0003] Vibration monitoring, temperature monitoring, oil analysis and acoustic monitoring are common methods for detecting helicopter gear faults. Vibration monitoring is easily disturbed by environmental noise and other mechanical components, and data analysis is relatively complex. It is easy to report errors, resulting in unnecessary maintenance costs. Temperature monitoring is usually a late symptom of a fault, which is difficult to prevent and respond to in a timely manner, and can easily lead to huge losses. The accuracy of the sensor may decrease in extreme environments, and a single sensor is difficult to fully monitor the temperature status of the entire system. Oil analysis usually requires regular sampling, and real-time monitoring cannot be achieved. The analysis equipment is complex and requires professional operation. Acoustic monitoring is easily disturbed by environmental noise, the complexity of signal processing, and has limitations in identifying more hidden internal faults. As a new detection method in the field of gear fault detection, electrical signal detection is suitable as a method for helicopter gear fault detection because of its simple signals, few interference factors, and comprehensive identification.
[0004] With the development of the Internet of Things and big data analysis technology, intelligent diagnosis and predictive maintenance based on deep learning are emerging detection methods. By uploading the data collected by sensors to the cloud platform and using machine learning and artificial intelligence algorithms for data analysis, real-time monitoring and fault prediction can be achieved. These systems can not only detect existing faults, but also predict possible future problems based on historical data and operating modes, thereby achieving preventive maintenance and minimizing downtime and maintenance costs. Intelligent diagnosis and predictive maintenance are still in their infancy, and practical applications are still relatively rare. They have great prospects as helicopter gear fault detection. Summary of the invention
[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides an online monitoring device and method for the lubrication status of gears based on deep learning, which can collect electrical signals on the surface of gears in real time, and use the collected electrical signals to identify the operating status of the gears. The neural network structure is simple and responsive, thereby realizing rapid online monitoring of gears. At the same time, a self-driving system is added to reduce the maintenance cycle and cost of the gear monitoring device.
[0006] To achieve the technical purpose of the present invention, the present invention will adopt the following technical solutions: A gear lubrication status online monitoring device based on deep learning, including a protective housing, a signal acquisition system, a self-power supply system and a self-monitoring control system, characterized in that;
[0007] The protective shell includes a seal, fixing screws, a first shell of the collection surface, a second shell of the collection surface, and a shell of the non-collection surface. The first shell of the collection surface is sleeved in the second shell, and there is a circular gap between the two shells.
[0008] The signal acquisition system comprises an electrode contact body, a base, a sponge body and an insulator, wherein the base, the sponge body and the insulator are sequentially sleeved on the electrode contact body and installed in the annular gap between the first shell of the acquisition surface and the second shell of the acquisition surface;
[0009] The self-powered system is installed on the inner wall of the protective housing;
[0010] The self-monitoring control system is wirelessly connected to the signal acquisition system. The electrical signal collected by the signal acquisition system is amplified and transmitted to the self-monitoring control system, and then the gear running status corresponding to the incoming electrical signal is analyzed based on the built-in program of the self-monitoring control system;
[0011] The first shell of the collection surface, the second shell of the collection surface and the shell of the non-collection surface respectively have four threaded blind holes and one arc groove, four threaded blind holes and one arc groove and eight threaded through holes and two arc grooves, wherein the first shell of the collection surface and the shell of the non-collection surface both have a through hole with a keyway in the middle;
[0012] The sealing member and the fixing screws are installed in 8 threaded holes and two arc grooves of the first outer shell of the collection surface, the second outer shell of the collection surface and the outer shell of the non-collection surface.
[0013] Furthermore, the first shell of the collection surface and the non-collection shell have the same card slots in the four directions of top, bottom, left and right and are centrally symmetrical.
[0014] Furthermore, the self-power supply system is installed in four identical card slots located at the top, bottom, left and right of the protective shell.
[0015] Furthermore, the electrode contact body can collect the potential of the tip, has a variety of collection frequencies and collection ranges, and has a circuit for amplifying electrical signals inside.
[0016] Furthermore, the online monitoring device and method for the lubrication status of gears as described in claim 1 is characterized in that the self-powered system is composed of a motor, a screw, a slider, a spring, a weight and a piezoelectric sheet from top to bottom, and the orientation of the self-powered system is changed by the rotation of the entire device, resulting in a change in the pressure of the heavy object on the piezoelectric sheet to generate piezoelectricity; the motor determines the size of the centripetal force exerted on the weight according to the size of the piezoelectricity generated by the detected piezoelectric sheet, and determines the size of the up and down movement of the slider and the spring by the rotation of the screw driven by the motor, thereby offsetting the centripetal force of the weight, so that the force of the weight on the piezoelectric sheet changes only due to the change in the orientation of the weight.
[0017] The invention also provides a monitoring method for a self-driving gear lubrication self-monitoring device, which is characterized by comprising the following steps:
[0018] Step 1: Install the self-driving gear lubrication self-monitoring device on the side of the gear close to the shaft shoulder;
[0019] Step 2: Open the debugging panel of the self-monitoring control system, and collect data of the gear under different operating conditions in the oil-rich state, oil-deficient state, and dry running state. The collection time for each state should be equal; after the collection is completed, click the training of the debugging panel to update the parameters of the fully connected neural network in the self-monitoring model;
[0020] Step 3: Turn on the self-monitoring mode, collect electrical signals from the gear surface, and import each 100 data points into the self-monitoring program as a group of data. The self-monitoring model outputs several set gear operating conditions.
[0021] Step 2 collects data under different operating conditions of the gear. The specific data collection processing method is as follows: each set of data contains 100 collection points, and the data starting points are 1, 49, 1, and 49 cycle collection points respectively; the fully connected neural network in step 2 is a six-layer neural network, with 80 neurons in the first layer, 60 neurons in the second layer, 40 neurons in the third layer, 20 neurons in the fourth layer, 10 neurons in the fifth layer, and 3 neurons in the sixth layer. The activation function used is the sigmoid activation function Sig(x)=(1+e -x ) -1 , the input layer and each layer of the neural network are connected by an activation function.
[0022] Furthermore, step 2 updates the parameters of the fully connected neural network in the self-monitoring model. The specific method is:
[0023] Step 2.1: Divide the data of gears under different operating conditions into 80% as training set and 20% as validation set;
[0024] Step 2.2: The input electrical signal is forward propagated in the neural network to output the probability of each result;
[0025] Step 2.3: Perform back propagation to update the parameters in the fully connected neural network based on the probability of the output result compared with the true value.
[0026] 9. The monitoring method of the gear lubrication state online monitoring device according to claim 6 is characterized in that the forward propagation in step 2.2 is specifically carried out by forward propagation through the formula h(l)=f(W(l)h(l-1)+b(l), where h is the output matrix of the lth layer, f is the activation function, W is the weight matrix of the lth layer, and b is the bias matrix of the lth layer.
[0027] Furthermore, step 2.3 back propagates, specifically through the gradient descent method To update the weight matrix of layer l, To update the bias matrix of the lth layer, where L is the loss function, a is the learning rate, It is L(W(l)) that finds the partial derivative of W(l).
[0028] The beneficial effects of the present invention are:
[0029] First, the present invention proposes a gear self-monitoring based on electrical signals, which has simpler data and faster response compared with conventional vibration, temperature and other detections.
[0030] Second, the present invention can monitor the lubrication status inside the gear in real time. By detecting the electrical signal on the gear surface, the collected data is analyzed using a deep learning neural network to obtain the lubrication status of the gear. This method is novel, low-cost, fast in response, and highly accurate.
[0031] Third, the present invention can change the training model according to the needs, and train the optimal model by collecting electrical signals under different states. It is applicable to most gears, and the output can be adjusted to monitor the state that needs to be monitored.
[0032] Fourth, the present invention adopts a self-monitoring method to monitor the gears, realizes online monitoring, reduces maintenance costs, avoids further development of gear failures, reduces unnecessary maintenance work, and extends the service life of the equipment.
[0033] Fifth, the present invention realizes self-driving of the device, which greatly reduces the maintenance cycle of the device and reduces the cost of detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a schematic diagram of the on-line monitoring device and method of gear lubrication status based on deep learning neural network of the present invention;
[0035] Figure 2It is an exploded diagram of the electrode system of the on-line monitoring device and method for gear lubrication status based on deep learning neural network of the present invention;
[0036] Figure 3 It is a schematic diagram of a self-powered system of an online monitoring device and method for gear lubrication status based on a deep learning neural network of the present invention;
[0037] Figure 4 The fully connected neural network structure of the on-line monitoring device and method for the gear lubrication state based on deep learning neural network of the present invention;
[0038] Figure 5 A potential diagram of dry operation collected by the on-line monitoring device and method for gear lubrication status based on deep learning neural network of the present invention;
[0039] Figure 6 The potential diagram of oil deficiency collected by the on-line monitoring device and method for gear lubrication status based on deep learning neural network of the present invention;
[0040] Figure 7 The oil-rich potential diagram collected by the on-line monitoring device and method for the lubrication state of gears based on deep learning neural network of the present invention;
[0041] Figure 8 The loss value and accuracy of the neural network model training process of the gear lubrication state online monitoring device and method based on deep learning neural network of the present invention;
[0042] Fig. 9 The electrical signals and input data under different states that need to be collected by the on-line monitoring device and method for the lubrication state of gears based on deep learning neural network of the present invention;
[0043] Fig.10 The results and accuracy of the training and test verification of the online monitoring device and method for gear lubrication status based on deep learning neural network of the present invention;
[0044] in, Figure 1 :1-protective shell; 2-signal acquisition system; 3-self-power supply system; 4-self-monitoring control system; 5-first shell of acquisition surface; 6-second shell of acquisition surface; 7-non-collection surface shell. Figure 2 :8-insulator; 9-sponge; 10-base; 11-electrode contact body: 2-signal acquisition system. Figure 3 :12-seal; 13 screw; 14-motor; 15-screw; 16-slider; 17-spring; 18-weight; 19-piezoelectric piece. DETAILED DESCRIPTION
[0045] In order to enable those skilled in the art to better understand the technical solution of the present application, the present application will be further described in detail below with reference to the accompanying drawings and embodiments.
[0046] The terms such as up, down, left, right, front and back in this application document are based on the positional relationships shown in the drawings. If the drawings are different, the corresponding positional relationships may also change accordingly, so they cannot be understood as limiting the scope of protection.
[0047] In this application, the terms "install", "connected", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, an integral connection, a mechanical connection, an electrical connection, or mutual communication, a direct connection, or an indirect connection through an intermediate medium, the internal connection of two components, or the interaction relationship between two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0048] Example 1
[0049] Refer to the instruction manual Figure 1 , a gear lubrication status online monitoring device based on deep learning, including a protective shell 1, a signal acquisition system 2, a self-power supply system 3 and a self-monitoring control system 4. The protective shell 1 is mainly composed of five parts, namely a seal 12, a fixing screw 13, a first shell 5 of the collection surface, a second shell 6 of the collection surface and a non-collection surface shell 7. The first shell 5 of the collection surface is sleeved in the second shell, and there is a circular gap between the two shells; the signal acquisition system 2 includes an electrode contact body 11, a base 10, a sponge 9 and an insulator 8. The base 10, the sponge 9 and the insulator 8 are sequentially sleeved on the electrode contact body and installed in the circular gap between the first shell 5 of the collection surface and the second shell 6 of the collection surface; a self-power supply system 3 is composed of a motor 14, a screw rod 15, a slider 16, a spring 17, a weight 18 and a piezoelectric sheet 19 from top to bottom, and is installed on the inner wall of the protective shell 1; the self-monitoring control system 4 is wirelessly connected to the signal acquisition system 2, and the electrical signal collected by the signal acquisition system 2 is amplified and transmitted to the self-monitoring control system 4, and then the state of the gear operation corresponding to the incoming electrical signal is analyzed based on the built-in program of the self-monitoring control system 4.
[0050] By adopting the structure of this embodiment, the electrical signal detection technology is combined with the deep learning neural network to realize online monitoring of gears, effectively solving the problem of complex gear maintenance and difficulty in real-time monitoring, reducing maintenance costs, and enabling operation and maintenance personnel to promptly discover and solve potential problems, avoid further development of faults, reduce unnecessary maintenance work, and extend the service life of equipment.
[0051] Example 2
[0052] On the basis of Example 1, the first shell of the collection surface and the second shell of the collection surface are respectively as follows Figure 1 The collecting surface first shell 5 and the collecting surface second shell 6 are shown.
[0053] Refer to the instruction manual Figure 2 The signal acquisition system 2 includes an electrode contact body 11, a base 10, a sponge body 9 and an insulator 8. The base 10, the sponge body 9 and the insulator 8 are sequentially sleeved on the electrode contact body and installed in the annular gap between the first shell 5 of the collection surface and the second shell 6 of the collection surface. When the gear and the gear lubrication state online monitoring device and method are installed closely, the insulator 8 is pressed downward and the sponge body 9 is compressed, and the electrode contact body 11 is exposed and contacts the gear, wherein the insulator 8 isolates the gear to prevent the gear oil from contacting the electrode insulator 11, and the base 10 is used to fix the electrode insulator 11. The electrode contact body 11 contacts the gear to collect weak electrical signals, and then amplifies the collected electrical signals through the amplification circuit and finally transmits them to the self-monitoring system.
[0054] By adopting the structure of this embodiment, the electrode insulator 11 is fixed by the base 10, and the insulator 8 and the sponge 9 are used to isolate interference, so as to achieve signal collection and reduce interference.
[0055] Example 3
[0056] On the basis of Example 1, the non-collecting surface shell, the collecting surface first shell and the collecting surface second shell are respectively as follows Figure 1 The non-collecting surface shell 7, the collecting surface first shell 5 and the collecting surface second shell 6 are shown.
[0057] Refer to the instruction manual Figure 3 The seal 12 is divided into two types, large and small. The small seal 12 is installed in the arc groove between the first shell 5 of the collection surface and the non-collection surface shell 7, and the large seal 12 is installed in the arc groove between the second shell 6 of the collection surface and the non-collection surface shell 7. The fixing screws 13 are installed in the 8 threaded holes of the first shell 5 of the collection surface, the second shell 6 of the collection surface and the non-collection surface shell 7.
[0058] Refer to the instruction manual Figure 3 The self-powered system 3 changes its orientation by rotating the entire device, causing the pressure of the weight 18 on the piezoelectric sheet 19 to change, thereby generating piezoelectricity. The motor 12 determines the magnitude of the centripetal force on the weight according to the magnitude of the piezoelectricity generated by the detected piezoelectric sheet 19, and determines the magnitude of the upward and downward movement of the slider 17 and the spring 17 by the rotation of the lead screw 15 driven by the motor 14, thereby offsetting the centripetal force of the weight 18, so that the force of the weight 18 on the piezoelectric sheet 19 changes only due to the change in the orientation of the weight 18.
[0059] By adopting the structure of this embodiment, the whole system is rotated to make the weight reciprocate, so that the piezoelectric film is subjected to reciprocating force to generate current, which supplies power to the motor 12 of the whole signal acquisition system 2 and the self-power supply system 3, thereby realizing self-driving.
[0060] Example 4
[0061] Refer to the instruction manual Figure 4 , Figure 5 and Fig.10 In combination with the above-mentioned embodiments 1-3, the present invention further provides a method for an online monitoring device for gear lubrication status based on deep learning, the method comprising the following steps;
[0062] Step 1: Install the self-driving gear lubrication self-monitoring device on the side of the gear close to the shaft shoulder;
[0063] Step 2: Open the debugging panel of the self-monitoring control system 4, and make the gear run under different conditions, such as: oil-rich state, oil-deficient state and dry running state, collect data under different running states of the gear, and the collection time of each state should be equal. After the collection is completed, click the training of the debugging panel to update the parameters of the fully connected neural network in the self-monitoring model;
[0064] The fully connected neural network is a six-layer neural network, with 80 neurons in the first layer, 60 neurons in the second layer, 40 neurons in the third layer, 20 neurons in the fourth layer, 10 neurons in the fifth layer, and 3 neurons in the sixth layer. The activation function used is the sigmoid activation function Sig(x)=(1+e -x ) -1 , the input layer and each layer of the neural network are connected by an activation function.
[0065] by Figure 4 Taking the figure as an example, X1, X2, X3, X4, X5, X6...X99, X100 are the input electrical signals, which pass through 5 layers of neural network and finally output Y1, Y2, Y3. The parameters in the updated neural network are W611W612... and b611b612... in the figure. There is a parameter matrix of W and b between every two layers of neural network.
[0066] Step 3: Turn on the self-monitoring mode, collect electrical signals from the gear surface, and import each 100 data points into the self-monitoring program as a group of data. The self-monitoring model outputs several set gear operating conditions.
[0067] by Fig.10 Taking the example shown in the figure, the predicted value is the corresponding label value set by the self-monitoring output result, and the true value is the gear lubrication state originally corresponding to the input data.
[0068] The data of the gear under different operating conditions are collected. The specific data processing method is as follows: each set of data contains 100 collection points, and the data starting points are respectively separated by 1, 49, 1, and 49 cycle collection points.
[0069] Step 2.1: Divide the data collected from gears under different operating conditions into 80% as a training set and 20% as a validation set.
[0070] Step 2.2: The input electrical signal is forward propagated in the neural network to output the probability of each result.
[0071] Step 2.3: Perform back propagation to update the parameters in the fully connected neural network based on the probability of the output result compared with the true value.
[0072] The forward propagation is carried out through the formula h(l)=f(W(l)h(l-1)+b(l), where h(l) is the output matrix of the lth layer, f is the activation function, W(l) is the weight matrix of the lth layer, and b(l) is the bias matrix of the lth layer; the backward propagation is carried out through the gradient descent method. To update the weight matrix of layer l, To update the bias matrix of the lth layer, where L is the loss function, a is the learning rate, L(W(l)) is the partial derivative of W(l); the loss function is L = -[ylog y^+(1-y)log(1-y^), where y is the true label value matrix and y^ is the predicted probability value matrix; the learning rate a is between 0.0001 and 0.002.
[0073] Figure 5 The loss value and accuracy of the neural network are shown during the process of updating the parameters of the fully connected neural network.
[0074] Example 5
[0075] Refer to the instruction manual Figure 5-7 In combination with the above-mentioned embodiment 4, the horizontal axis is time, the vertical axis is voltage, the voltage is the potential difference between the gear surface potential and the ground potential, and the gear surface voltage oscillates back and forth, which can be divided into two sections: positive voltage and negative voltage. Figure 5 This is the voltage diagram detected when the gear is running dry. The difference between the peak and the trough of the positive and negative voltages is large. The variance of the positive voltage is 1.277, and the variance of the negative voltage is 1.473, that is, the voltage data fluctuates greatly. Figure 6 This is the potential diagram detected when the gear is short of oil. The difference between the peak and the trough of the positive and negative voltages is small. The variance of the positive voltage is 0.478, and the variance of the negative voltage is 0.511, that is, the fluctuation of the electrical signal is small. Figure 7This is the potential diagram detected when the gear is in the oil-rich state. There are no obvious peaks and valleys in the positive and negative voltages. The variance of the positive voltage is 0.132, and the variance of the negative voltage is 0.314, that is, the electrical signal fluctuates weakly. There are obvious differences in the potential diagrams under the three operating states of the gear. The potential diagram features of the dry running state, the oil-deficient state, and the oil-rich state are different. The fully connected neural network updates parameters through multiple learning to extract the features of the potential diagrams under various states and output the corresponding correct results. Similarly, there are obvious differences in the electrical signals detected when the gear is overloaded, normally loaded, and unloaded. That is, the fully connected neural network learns and updates parameters to extract the features of the potential diagrams under various states to achieve the desired monitoring function.
[0076] Example 6
[0077] Refer to the instruction manual Figure 4 and Fig. 9 , combined with the above-mentioned Example 3 and Example 4, Fig. 9 The first column is the data group number and does not participate in training. The second to 101 columns are data group data. One row is a group of data. The corresponding numbers X1, X2, X3, X4, X5, X6...X98, X99, X100 are imported respectively. Figure 4 In the corresponding input neuron, the 102nd column Label is a label. In this embodiment, 0 represents dry operation, 1 represents oil-rich operation, and 2 represents oil-poor operation.
[0078] Example 1 data, the gear is in dry running state, Figure 5 Excluding the data in the first row of the header row, the data in the columns X1, X2, X3, X4, X5, X6...X98, X99, and X100 are imported into the neural network. After the first layer of the neural network sigmoid activation function and the calculation of the W1 b1 matrix, the data of 80 neurons are similar to X1, X2, X3, X4, X5, X6...X98, X99, and X100. Then the data of 60 neurons are obtained after the second layer of the neural network sigmoid activation function and the W2 b2 matrix. Then the data of 40 neurons are obtained after the third layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Then the data of 20 neurons are obtained after the fourth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Then the data of 10 neurons are obtained after the fifth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Finally, the data of 3 neurons are obtained after the sixth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Figure 4Y1, Y2, and Y3. In this embodiment, Y1 is the probability value of label 0, Y2 is the probability value of label 1, and Y3 is the probability value of label 2. The range of Y1, Y2, and Y3 is 0 to 1. At this time, the outputs Y1, Y2, and Y3 should ideally be 1, 0, and 0, but there are actual deviations. Y1 is a number close to 1 and less than 1, Y2 is a number close to 0 and greater than 0, and Y3 is a number close to 0 and greater than 0. The label corresponding to the label value with the largest probability value output by the model is 0, and the label corresponding to the label value with the largest probability value is "dry operation".
[0079] In case 3, the gear is in a state of lack of oil. Figure 5 Excluding the data in the third row of the title row, the data in the columns X1, X2, X3, X4, X5, X6...X98, X99, and X100 are imported into the neural network. After the first layer of the neural network sigmoid activation function and the calculation of the W1 b1 matrix, the data of 80 neurons are similar to X1, X2, X3, X4, X5, X6...X98, X99, and X100. Then the data of 60 neurons are obtained after the second layer of the neural network sigmoid activation function and the W2 b2 matrix. Then the data of 40 neurons are obtained after the third layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Then the data of 20 neurons are obtained after the fourth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Then the data of 10 neurons are obtained after the fifth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Finally, the data of 3 neurons are obtained after the sixth layer of the sigmoid activation function and the W3 b3 matrix of the neural network. Figure 4 Y1, Y2, and Y3. In this embodiment, Y1 is the probability value of label 0, Y2 is the probability value of label 1, and Y3 is the probability value of label 2. The range of Y1, Y2, and Y3 is 0 to 1. At this time, the outputs Y1, Y2, and Y3 should ideally be 0, 1, and 0, but there are actual deviations. Y1 is a number close to 0 and greater than 0, Y2 is a number close to 1 and less than 1, and Y3 is a number close to 0 and greater than 0. The model outputs the label value corresponding to the largest probability value, and the corresponding maximum label value is 1. The label corresponding to the label value is "lack of oil".
[0080] In case 5, the gear is in the oil-rich state. Figure 5Excluding the header row, the data in the 5th row X1, X2, X3, X4, X5, X6...X98, X99, X100 columns are imported into the neural network. After the first layer of neural network sigmoid activation function and W1 b1 matrix calculation, the data of 80 neurons are similar to X1, X2, X3, X4, X5, X6...X98, X99, X100. Then the data of 60 neurons are obtained after the second layer of neural network sigmoid activation function and W2 b2 matrix. Then the data of 40 neurons are obtained after the third layer of sigmoid activation function and neural network W3 b3 matrix. Then the data of 20 neurons are obtained after the fourth layer of sigmoid activation function and neural network W3 b3 matrix. Then the data of 10 neurons are obtained after the fifth layer of sigmoid activation function and neural network W3 b3 matrix. Finally, the data of 3 neurons are obtained after the sixth layer of sigmoid activation function and neural network W3 b3 matrix. Figure 4 Y1, Y2, and Y3. In this embodiment, Y1 is the probability value of label 0, Y2 is the probability value of label 1, and Y3 is the probability value of label 2. The range of Y1, Y2, and Y3 is 0 to 1. At this time, the outputs Y1, Y2, and Y3 should ideally be 0, 0, and 1, but there are actual deviations. Y1 is a number close to 0 and greater than 0, Y2 is a number close to 0 and greater than 0, and Y3 is a number close to 1 and less than 1. The model outputs the label value corresponding to the maximum probability value, and the corresponding label value is 2. The label corresponding to the maximum label value is "rich in oil".
[0081] The above is only an embodiment of the present invention, and the common sense such as the known specific structure and characteristics in the scheme is not described in detail here. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the attached claims rather than the above description, so it is intended to include all changes within the meaning and scope of the equivalent elements of the claims. Any figure mark in the claims should not be regarded as limiting the claims involved.
Claims
1. A gear lubrication status online monitoring device based on deep learning, comprising a protective housing (1), a signal acquisition system (2), a self-power supply system (3) and a self-monitoring control system (4), characterized in that ; The protective housing (1) comprises a sealing member (12), a fixing screw (13), a first housing (5) of the collection surface, a second housing (6) of the collection surface and a housing (7) of the non-collection surface. The first housing (5) of the collection surface is sleeved in the second housing, and a circular gap exists between the two housings. The signal acquisition system (2) comprises an electrode contact body (11), a base (10), a sponge body (9) and an insulator (8); the base (10), the sponge body (9) and the insulator (8) are sequentially sleeved on the electrode contact body (11) and installed in the annular gap between the first shell (5) of the acquisition surface and the second shell (6) of the acquisition surface; The self-powered system (3) is installed on the inner wall of the protective housing (1); The self-monitoring control system (4) is wirelessly connected to the signal acquisition system (2), and the electrical signal collected by the signal acquisition system (2) is amplified and transmitted to the self-monitoring control system (4), and then the gear operation state corresponding to the transmitted electrical signal is analyzed based on the built-in program of the self-monitoring control system (4); The first shell (5) of the collection surface, the second shell (6) of the collection surface and the shell (7) of the non-collection surface respectively have four threaded blind holes and one arc groove, four threaded blind holes and one arc groove and eight threaded through holes and two arc grooves, wherein the first shell (5) of the collection surface and the shell (7) of the non-collection surface both have a through hole with a keyway in the middle; The sealing member (12) and the fixing screw (13) are installed in eight threaded holes and two arc-shaped grooves of the first collection surface shell (5), the second collection surface shell (6) and the non-collection surface shell (7).
2. The on-line monitoring device for gear lubrication status according to claim 1 is characterized in that: The first shell (5) of the collection surface and the non-collection shell (7) have identical card slots in four directions, namely, directly above, directly below, directly to the left and directly to the right, and are centrally symmetrical.
3. The on-line monitoring device and method for gear lubrication status according to claim 1, characterized in that: The self-power supply system (3) is installed in four identical card slots located at the top, bottom, left and right of the protective housing (1).
4. The on-line monitoring device for gear lubrication status according to claim 1, characterized in that: The electrode contact body (11) can collect the potential of the tip, has a variety of collection frequencies and collection ranges, and has a circuit for amplifying electrical signals inside.
5. The on-line monitoring device for gear lubrication status according to claim 1, characterized in that: The self-powered system (3) comprises, from top to bottom, a motor (14), a screw rod (15), a slider (16), a spring (17), a weight (18) and a piezoelectric sheet (19). The self-powered system (3) changes its orientation by rotating the entire device, causing the pressure of the weight (18) on the piezoelectric sheet (19) to change, thereby generating piezoelectricity. The motor (12) determines the size of the centripetal force on the weight according to the size of the piezoelectricity generated by the detected piezoelectric sheet (19), and determines the size of the upward and downward movement of the slider (17) and the spring (17) by the motor (14) driving the screw rod (15) to rotate, thereby offsetting the centripetal force of the weight (18), so that the force of the weight (18) on the piezoelectric sheet (19) changes only due to the change in the orientation of the weight (18).
6. A monitoring method for the self-driving gear lubrication self-monitoring device according to claims 1-5, characterized in that: The following steps are involved: Step 1: Install the self-driving gear lubrication self-monitoring device on the side of the gear close to the shaft shoulder; Step 2: Open the debugging panel of the self-monitoring control system (4), and collect data of the gear under different operating conditions in the oil-rich state, the oil-deficient state, and the dry running state. The collection time of each state should be equal; After the acquisition is completed, click the training of the debugging panel to update the parameters of the fully connected neural network in the self-monitoring model; Step 3: Turn on the self-monitoring mode, collect electrical signals from the gear surface, and import each 100 data points into the self-monitoring program as a group of data. The self-monitoring model outputs several set gear operating conditions.
7. The monitoring method of the gear lubrication status online monitoring device according to claim 6, characterized in that: Step 2 collects data under different operating conditions of the gear. The specific data collection processing method is as follows: each set of data contains 100 collection points, and the data starting points are 1, 49, 1, and 49 cycle collection points respectively; the fully connected neural network in step 2 is a six-layer neural network, with 80 neurons in the first layer, 60 neurons in the second layer, 40 neurons in the third layer, 20 neurons in the fourth layer, 10 neurons in the fifth layer, and 3 neurons in the sixth layer. The activation function used is the sigmoid activation function Sig(x)=(1+e -x ) -1 , the input layer and each layer of the neural network are connected by an activation function.
8. The monitoring method of the gear lubrication status online monitoring device according to claim 6, characterized in that: Step 2 updates the parameters of the fully connected neural network in the self-monitoring model. The specific method is: Step 2.1: Divide the data of gears under different operating conditions into 80% as training set and 20% as validation set; Step 2.2: The input electrical signal is forward propagated in the neural network to output the probability of each result; Step 2.3: Perform back propagation to update the parameters in the fully connected neural network based on the probability of the output result compared with the true value.
9. The monitoring method of the gear lubrication status online monitoring device according to claim 6, characterized in that: Step 2.2 forward propagation, specifically forward propagation through the formula h(l)=f(W(l)h(l-1)+b(l), where h(l) is the output matrix of the first layer, f is the activation function, W(l) is the weight matrix of the first layer, and b(l) is the bias matrix of the first layer.
10. The monitoring method of the gear lubrication status online monitoring device according to claim 6, characterized in that: Step 2.3 Back propagation, specifically through gradient descent To update the weight matrix of layer 1, To update the bias matrix of the first layer, where L is the loss function, a is the learning rate, It is L(W(l)) that finds the partial derivative of W(l).
