Crusher health monitoring device and method

Abstract

The invention relates to an equipment fault prediction technology, discloses a crusher health monitoring device and method, and solves the problem that a crusher health monitoring scheme in the prior art is unfavorable for comprehensively, accurately and efficiently mastering the health state of a crusher. According to the invention, the MEMS sensing technology is utilized, a temperature and humidity sensor, an air pressure sensor and a vibration acceleration sensor are integrated in the monitoring device, vibration parameters in the working process of the crusher are measured, various environmental parameters are also measured, and the data are used for performing multi-modal fusion with power spectrum data estimated according to vibration signals of the crusher, so that the monitoring accuracy is improved. A multi-parameter fusion feature vector is constructed, so that more comprehensive and accurate data information is provided; and in the aspect of health state prediction, working condition prediction based on multi-parameter fusion feature vectors is realized by training a BP neural network model, so that the health state of the crusher is mastered more comprehensively, accurately and efficiently. The method is suitable for fault prediction of the crusher.

Description

Technical Field

[0001] The present invention relates to equipment failure prediction technology, and in particular to a crusher health monitoring device and method. Background Art

[0002] Crusher is mainly used in industrial scenarios such as ore mining and processing, waste material recycling, etc. In the lithium battery recycling industry, the use of crusher for mechanical processing can increase scale, reduce costs, and achieve the best recycling and separation of waste lithium batteries. Therefore, crusher is an important part of the lithium battery recycling and sorting line. In order to reduce the difficulty of management, enterprises need to incorporate the crusher equipment of the production line into the enterprise-level monitoring system for management, monitor the operating status and working conditions of the equipment in real time, and realize data aggregation analysis and prediction.

[0003] The existing technology is often limited to the measurement and acquisition of the operating parameters of the crusher itself, and after the data is feature extracted, a set threshold is used to determine or predict whether the equipment is faulty. Under actual complex working conditions, different equipment has different fault thresholds, so the application of this solution is often limited, and the analysis and prediction are not accurate and timely. For example: The patent with publication number CN118961183A provides a fault monitoring method and system for crushing and screening equipment, which determines whether there is a fault by judging whether the rotation speed exceeds the stable range and the linear corresponding deviation value and asynchronous response value of each monitoring data in the divided time interval. Since this method mainly collects the dynamic parameters of the crusher itself during operation, such as the rotation speed, the parameters of the working environment are not collected, and the measurement of the working condition parameters requires the arrangement of additional equipment and manpower to obtain through patrol inspections or regular inspections, it is not conducive to comprehensively, accurately and efficiently grasping the health status of the crusher.

[0004] As the online monitoring system of the enterprise intelligent information platform develops towards multimodal deep learning, there is an urgent need for a device and method that can measure the operating vibration parameters of crusher equipment and various working environment parameters to provide more comprehensive and accurate data information and modal analysis. Summary of the invention

[0005] The technical problem to be solved by the present invention is to propose a crusher health monitoring device and method to solve the problem that the existing crusher health monitoring scheme is not conducive to comprehensively, accurately and efficiently grasping the health status of the crusher.

[0006] The technical solution adopted by the present invention to solve the above technical problems is:

[0007] In one aspect, the present invention provides a crusher health monitoring device, comprising: a housing, a small plate, a main plate and a cover plate;

[0008] The shell is provided with a through bolt hole, the side of the shell is provided with a cable hole, the top of the shell is provided with a sinking pit inside, and the center of the sinking pit is provided with a through hole; the inner wall of the shell is provided with a cutting edge near the bottom;

[0009] The small board is installed in the sunken pit of the housing, with a temperature and humidity sensor and an air pressure sensor integrated on the front and a cable socket integrated on the back;

[0010] The mainboard is mounted on the cutting edge of the inner wall of the housing and encapsulates an MCU, an acceleration sensor and a cable socket;

[0011] The cover plate is fixed to the bottom of the shell, and bolt holes corresponding to the bolt holes of the shell are arranged on the cover plate.

[0012] Furthermore, the shell is integrally formed of stainless steel.

[0013] Furthermore, the small board is made of PCB material, and the temperature and humidity sensor and air pressure sensor integrated on the front are exposed outward through the through hole in the center of the sunken pit of the shell, and the cable socket on the back is connected to the cable socket on the mainboard.

[0014] Furthermore, the mainboard is made of PCB material, and a semicircular notch is provided on its edge.

[0015] Furthermore, the mainboard is provided with pin solder joints for realizing power supply and signal output through cables; the cables pass through the cable holes on the housing.

[0016] Furthermore, the cover plate is made of stainless steel and is laser welded to the bottom of the housing, and the bolt holes on the cover plate correspond to the bolt holes on the housing.

[0017] Furthermore, a small board and a main board are installed inside the shell, and after wiring, epoxy resin glue is poured through the semicircular notch of the main board and cured.

[0018] On the other hand, the present invention also provides a crusher health monitoring method, comprising:

[0019] S1. Construct training data samples:

[0020] S11, according to the vibration signal of the crusher, the power spectrum is estimated by using an AR (autoregressive) model;

[0021] S12, normalizing the estimated power spectrum with the temperature, humidity, air pressure and vibration acceleration data collected at the vibration signal time point of the crusher to construct a multi-parameter fusion feature vector;

[0022] S13, constructing multi-parameter fusion feature vectors of the device under test in different operating modes and working conditions according to steps S11 and S12 as training data samples;

[0023] S2. Training health monitoring model:

[0024] Taking the training data samples as input and the corresponding operation modes and working conditions of the training data samples as output, the BP neural network model is trained by using the back propagation and gradient descent algorithms until the training termination condition is reached, and the trained BP neural network model is obtained as the health monitoring model;

[0025] S3. Perform health monitoring tasks:

[0026] Collect the current vibration signal, current temperature, current humidity, current air pressure, and current vibration acceleration of the equipment to be tested; estimate the current power spectrum based on the current vibration signal, normalize the current power spectrum with the current temperature, current humidity, current air pressure, and current vibration acceleration, and construct a multi-parameter fusion feature vector as the input of the health monitoring model, and output the current operating mode and working condition of the crusher based on the health monitoring model.

[0027] Furthermore, in step S11, the power spectrum is estimated using an AR model according to the vibration signal of the crusher, including:

[0028] Calculate forward prediction error and backward prediction error;

[0029] Calculating a reflection coefficient based on the forward prediction error and the backward prediction error;

[0030] Calculate the autoregressive coefficient and mean square error according to the reflection coefficient;

[0031] The power spectrum of the vibration signal is estimated based on the autoregressive coefficient and the mean square error.

[0032] The beneficial effects of the present invention are:

[0033] The present invention performs composite measurement of multiple parameters based on MEMS (micro-electromechanical system) sensor technology, and has the advantages of small size, low power consumption and high sensitivity. It also utilizes a multi-modal fusion algorithm to fuse various parameters and construct a multi-parameter fusion feature vector, which is used to train a BP neural network model to predict the operating condition of the crusher. The present invention can be used for health status monitoring and multi-modal parameter analysis of operating conditions of crusher equipment operating in outdoor complex environments or closed or semi-closed microenvironments, thereby more comprehensively, accurately and efficiently grasping the health status of the crusher. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 An exploded diagram of the structure of a crusher health monitoring device in an embodiment of the present invention;

[0035] Figure 2 Schematic diagram of the housing structure of the crusher health monitoring device in an embodiment of the present invention;

[0036] Figure 3 Schematic diagram of the crusher health monitoring method in an embodiment of the present invention;

[0037] Markings in the figure: 1 is the housing, 101 is the cable hole, 102 is the bolt hole, 103 is the sinking pit, 104 is the cutting edge, 2 is the small board, 3 is the main board, and 4 is the cover plate. DETAILED DESCRIPTION

[0038] The present invention aims to provide a crusher health monitoring device and method, so as to solve the problem that the existing crusher health monitoring scheme is not conducive to comprehensively, accurately and efficiently grasping the health status of the crusher. The core idea is: using MEMS sensing technology, integrating temperature and humidity, air pressure, and vibration acceleration sensors in the monitoring device, and then in addition to measuring the vibration parameters of the crusher during operation, various environmental parameters are also measured. These data are used for multi-modal fusion with the power spectrum data estimated from the crusher vibration signal to construct a multi-parameter fusion feature vector, so as to provide more comprehensive and accurate data information; and in terms of health status prediction, the BP neural network model is trained to realize the working condition prediction based on the multi-parameter fusion feature vector. Compared with the traditional threshold judgment method, it has stronger adaptability and learning ability in complex working conditions and can improve the accuracy of prediction.

[0039] The scheme of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0040] Example

[0041] See also Figure 1 The present embodiment provides a crusher health monitoring device, which includes a housing 1, a small plate 2, a main plate 3 and a cover plate 4; in specific implementation, the housing 1 is made of stainless steel and is integrally formed without a bottom; a through bolt hole 102 is provided on the housing 1, a cable hole 101 is provided on the side of the housing 1, a sinking pit 103 is provided inside the top of the housing 1, and a through hole is provided in the center of the sinking pit 103; a cutting edge 104 is provided near the bottom of the inner wall of the housing 1, see Figure 2 .

[0042] The small board 2 is made of PCB material and is installed in the sunken pit 103 of the housing 1. The front side of the small board 2 is integrated with a temperature and humidity sensor and an air pressure sensor, and the back side is integrated with a cable socket. The sensor integrated on the front side is exposed to the outside through the through hole on the lower pit 103 of the housing 1, and the back side is connected to the cable socket on the main board 3 through the cable socket. The small board 3 is treated with three protections except for the sensor and the cable socket. In addition, a detachable rubber cap can be provided at the through hole on the lower pit 103 of the housing 1 to protect the sensor, which can be removed when in use.

[0043] The mainboard 3 is made of PCB material, and the mainboard 3 is installed on the cutting edge 104 of the inner wall of the housing 1, which encapsulates the MCU, the acceleration sensor and the cable socket; the edge of the mainboard 3 has a semicircular notch, and the circuit of the mainboard 3 has pin solder joints, providing a DC regulated power supply port and an RS485 serial communication pin. After the pin part is welded with cables, the mainboard 3 is treated with three protections except for the cable socket part. The cable passes through the cable hole 101 on the housing 1 to achieve power supply and signal output.

[0044] After the small board 2 and the main board 3 are installed and connected inside the housing 1, epoxy resin glue is poured through the semicircular arc notch of the main board 3 and cured.

[0045] The cover plate 4 is made of stainless steel and is laser welded to the bottom of the housing 1 , and the bolt holes on the cover plate 4 correspond to the bolt holes 102 on the housing 1 .

[0046] Based on the above device, the vibration parameters of the crusher during operation as well as environmental parameters such as temperature, humidity, and air pressure can be collected.

[0047] In the crusher health monitoring method provided in this embodiment, its implementation principle is shown in Figure 3 Specifically, according to the vibration signal of the crusher, the AR model is used to calculate the autoregressive coefficient to estimate the power spectrum; the estimated power spectrum is normalized with the temperature, humidity, air pressure and vibration acceleration data collected at the time point of the vibration signal of the crusher to construct a multi-parameter fusion feature vector; by constructing the multi-parameter fusion feature vector of the equipment under test in different operating modes and working conditions, the BP neural network model is trained as a training data sample to obtain a health monitoring model.

[0048] Among them, the form of AR model is:

[0049]

[0050] In the formula, x(n) is the original signal sequence, a k is the autoregressive coefficient, u(n) is the white noise error term, and p is the model order.

[0051] The formula for estimating the power spectrum is:

[0052]

[0053] where σ 2 is the variance of white noise.

[0054] The forward prediction error of the model is expressed as:

[0055]

[0056] The backward prediction error of the model is expressed as:

[0057]

[0058] The reflection coefficient calculation formula is:

[0059]

[0060] The forward prediction error calculation formula is:

[0061] ef k (n) = ef k-1 (n)+γ k eb k-1 (n-1)

[0062] The backward prediction error calculation formula is:

[0063] eb k (n) = eb k-1 (n-1)+γ k ef k-1 (n)

[0064] The initial forward and backward prediction errors are:

[0065] ef0(n)=eb0(n)=x(n)

[0066] The initial mean square error is:

[0067]

[0068] By taking ef0(n), eb0(n), σ0 2 The reflection coefficient γ can be obtained by iterating into the reflection coefficient calculation formula and the front and back prediction error calculation formula k .

[0069] Calculate the autoregressive coefficient from the reflection coefficient:

[0070] a k (i) = a k-1 (i)+γ k a k-1 (ki)

[0071] Calculate the mean square error from the reflection coefficient:

[0072] σ k 2 =σ k-1 2 (1-γ k 2 )

[0073] Finally, substituting the autoregressive coefficient and the mean square error into the estimated power spectrum formula, we can get the estimated power spectrum P of the signal. x (e j ω ).

[0074] After obtaining the estimated power spectrum P x (e jω ) and then compare it with the temperature T measured by the sensor on the monitoring device i 、Humidity i , air pressure P i And the time domain vibration acceleration signal A after filtering is used to construct a multi-parameter feature vector:

[0075] X=[T,H,P,P(e jω ),A]

[0076] By collecting the multi-parameter fusion feature vector X of the device under test in different operating modes and working conditions as the input of the BP neural network model, and defining different operating modes and working conditions as Y as the output, the number of hidden layer nodes in the BP neural network can be determined by the empirical formula:

[0077]

[0078] Where n is the number of input X, m is the number of output Y, and a is a constant between 0 and 10.

[0079] The model training effects vary depending on the different devices to be tested and the working environment. The activation function can be selected based on the actual training effect, and the input samples are normalized before model training.

[0080] The training process of BP neural network includes: forward propagation and back propagation process; more specifically:

[0081] During the forward propagation process, the input data is linearly combined through the weight matrix and the bias vector, and then converted into a nonlinear expression through the activation function:

[0082]

[0083] Among them, α i is the hidden layer input, β k is the output layer input, f is the activation function, wij 、v ik is the weight, b i 、c k is the bias parameter.

[0084] The loss function of the back propagation process is:

[0085]

[0086] in, is the predicted output value, y i is the actual output value, E i is the output error.

[0087] The optimal solution is sought by the gradient descent method, and the error gradient is:

[0088]

[0089] Then update the weight matrix and bias parameters:

[0090]

[0091] Based on the above training process, on the training samples, after a certain number of training iterations, a converged BP neural network model can be obtained, and the preset parameters can be adjusted according to the model effect to finally obtain a health monitoring model.

[0092] In practical applications, the health monitoring task can be performed by deploying the health monitoring model. The execution process includes: collecting the current vibration signal, current temperature, current humidity, current air pressure, and current vibration acceleration of the equipment to be detected; estimating the current power spectrum based on the current vibration signal, normalizing the current power spectrum with the current temperature, current humidity, current air pressure, and current vibration acceleration, and constructing a multi-parameter fusion feature vector as the input of the health monitoring model, and outputting the current operating mode and working condition of the crusher based on the health monitoring model.

[0093] Finally, it should be noted that the above embodiments are only preferred implementations and are not intended to limit the present invention. It should be pointed out that for those skilled in the art, several modifications, equivalent replacements, improvements, etc. can be made without departing from the scope of the present invention and the scope of protection of the claims, and all of these should be included in the protection scope of the present invention.

Claims

1. A crusher health monitoring device, characterized in that: include: Housing, small board, main board and cover; The shell is provided with a through bolt hole, the side of the shell is provided with a cable hole, the top of the shell is provided with a sinking pit inside, and the center of the sinking pit is provided with a through hole; the inner wall of the shell is provided with a cutting edge near the bottom; The small board is installed in the sunken pit of the housing, with a temperature and humidity sensor and an air pressure sensor integrated on the front and a cable socket integrated on the back; The mainboard is mounted on the cutting edge of the inner wall of the housing and encapsulates an MCU, an acceleration sensor and a cable socket; The cover plate is fixed to the bottom of the shell, and bolt holes corresponding to the bolt holes of the shell are arranged on the cover plate.

2. A crusher health monitoring device according to claim 1, characterized in that: The shell is integrally formed of stainless steel.

3. A crusher health monitoring device according to claim 1, characterized in that: The small board is made of PCB material, and the temperature and humidity sensor and air pressure sensor integrated on the front are exposed outward through the through hole in the center of the sunken pit of the outer shell, and the cable socket on the back is connected to the cable socket on the mainboard.

4. A crusher health monitoring device according to claim 3, characterized in that: The main board is made of PCB material, and a semicircular notch is arranged on its edge.

5. A crusher health monitoring device according to claim 4, characterized in that: The mainboard is provided with pin soldering points for realizing power supply and signal output through cables; the cables are passed through the cable holes on the housing.

6. A crusher health monitoring device according to claim 1, characterized in that: The cover plate is made of stainless steel and is welded to the bottom of the shell by laser, and the bolt holes on the cover plate correspond to the bolt holes on the shell.

7. A crusher health monitoring device according to claim 4, characterized in that: The small board and the main board are installed inside the shell, and after the wiring is completed, epoxy resin glue is poured through the semicircular arc notch of the main board and cured.

8. A crusher health monitoring method, characterized in that: include: S1. Construct training data samples: S11, according to the vibration signal of the crusher, the power spectrum is estimated by using the AR autoregressive model; S12, normalizing the estimated power spectrum with the temperature, humidity, air pressure and vibration acceleration data collected at the vibration signal time point of the crusher to construct a multi-parameter fusion feature vector; S13, constructing multi-parameter fusion feature vectors of the device under test in different operating modes and working conditions according to steps S11 and S12 as training data samples; S2. Training health monitoring model: Taking the training data samples as input and the corresponding operation modes and working conditions of the training data samples as output, the BP neural network model is trained by using the back propagation and gradient descent algorithms until the training termination condition is reached, and the trained BP neural network model is obtained as the health monitoring model; S3. Perform health monitoring tasks: Collect the current vibration signal, current temperature, current humidity, current air pressure, and current vibration acceleration of the equipment to be tested; estimate the current power spectrum based on the current vibration signal, normalize the current power spectrum with the current temperature, current humidity, current air pressure, and current vibration acceleration, and construct a multi-parameter fusion feature vector as the input of the health monitoring model, and output the current operating mode and working condition of the crusher based on the health monitoring model.

9. A crusher health monitoring method according to claim 8, characterized in that: In step S11, the power spectrum is estimated using an AR model according to the vibration signal of the crusher, including: Calculate forward prediction error and backward prediction error; Calculating a reflection coefficient based on the forward prediction error and the backward prediction error; Calculate the autoregressive coefficient and mean square error according to the reflection coefficient; The power spectrum of the vibration signal is estimated based on the autoregressive coefficient and the mean square error.

Images