Online intelligent regulation and control method and system for gear life of vertical mill gearbox

Abstract

The invention belongs to the technical field of vertical mill gearboxes, and particularly discloses a vertical mill gearbox gear life online intelligent regulation and control method and system.The method comprises the steps that a main test gearbox gear rotating speed signal and a box body vibration acceleration signal are collected, whether a gear enters a degeneration stage or not is judged, and if the gear enters the degeneration stage, the gear is started; if yes, constructing a life prediction model and predicting the remaining life of the gear of the vertical mill gearbox by using the life prediction model; in combination with a vertical mill gearbox gear remaining life prediction result and the current expected remaining life, optimal regulation and control working condition parameters are calculated on the basis of a model-free adaptive regulation and control MFAC theory and by introducing a pseudo partial derivative PPD and a tight format dynamic linearization CFDL model; and based on the optimal regulation and control working condition parameters, the operation working condition of the vertical mill gearbox gear is regulated and controlled. By adopting the technical scheme, the remaining life of the gear of the vertical mill gearbox is predicted, the running state of the gear of the gearbox is regulated in real time, and the actual service life of the gear of the vertical mill gearbox is prolonged to be close to the expected service life.

Description

Technical Field

[0001] The invention belongs to the technical field of vertical mill gearboxes, and relates to an online intelligent control method and system for the life of gears in a vertical mill gearbox. Background Art

[0002] The vertical mill gearbox is one of the core components of power transmission equipment in industries such as cement and mining. Its operating stability is directly related to the efficiency and safety of the production line. However, due to its complex working environment, large load changes, and the gears and bearings in the gearbox being in harsh conditions such as high temperature, high humidity, and vibration for a long time, it is easy to cause wear, failure, and even major accidents.

[0003] Therefore, conducting research on the online intelligent control system for the gear life of the vertical mill gearbox is of great significance for extending the remaining service life of the vertical mill gearbox gears and improving the reliability and stability of the transmission system.

[0004] In the actual application of vertical mill equipment such as cement and mining, it is often possible to reasonably adjust the load of the equipment and optimize the operating conditions of the gears without affecting the normal functions of the equipment, thereby effectively extending its actual service life.

[0005] However, since most traditional gear life prediction systems remain at the remaining life prediction stage, they are unable to extend their actual service life to the expected life and cannot achieve the purpose of intelligent mechanical operation and maintenance. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for online intelligent control of the life of gears in a vertical mill gearbox, so as to realize real-time control of the operating state of the gears in the gearbox and extend the actual service life of the gears in the vertical mill gearbox to close to the expected service life.

[0007] In order to achieve the above object, the basic scheme of the present invention is: a method for online intelligent control of the gear life of a vertical mill gearbox, comprising the following steps:

[0008] Collect the gear speed signal and the vibration acceleration signal of the main test gearbox for 1s (interval) every 5 minutes;

[0009] Based on the 3σ criterion, determine whether the gear has entered the degradation stage. If the gear has entered the degradation stage, build a life prediction model and use the life prediction model to predict the remaining life of the vertical mill gearbox gear;

[0010] Combining the remaining life prediction results of the vertical mill gearbox gear and the current expected remaining life, the optimal control torque is calculated based on the model-free adaptive control MFAC theory and the introduction of pseudo partial derivative PPD and compact dynamic linearization CFDL model;

[0011] Based on the optimal control operating condition parameters, the operating conditions of the vertical mill gearbox gears are adjusted.

[0012] The working principle and beneficial effects of this basic solution are as follows: This technical solution collects the gear speed signal of the main test gearbox and the box vibration acceleration signal, and monitors in real time whether the gear has entered the degradation stage based on the 3σ criterion. When the gear enters the degradation stage, the remaining service life of the vertical mill gearbox gear is predicted, and then based on the remaining life prediction result and the current expected remaining life, the optimal control condition parameters are output. Accurate prediction and online intelligent control of the remaining service life of the vertical mill gearbox gear is achieved, and the actual service life of the vertical mill gearbox gear is successfully extended to close to the expected service life, ensuring the working efficiency and long-term stability of the gear transmission system, and reducing the economic losses and safety risks caused by accidental damage to the gear.

[0013] Furthermore, the life prediction model is constructed as follows:

[0014] S1, select the statistical features with the best trend and monotonicity of the gear speed signal and the vibration acceleration signal of the main gearbox in the whole life experiment as the health index of the gear, form a health index sequence in the whole life experiment, and construct the life sample on the health index sequence by sliding window method with a window length of 512 and a step size of 1;

[0015] S2, constructing a multivariate fusion feature extractor, extracting multivariate feature information related to degradation state and remaining service life in the life samples in parallel, fusing the multivariate feature information, and generating a final output of the multivariate fusion feature extractor;

[0016] S3, constructing a prediction integration module, the prediction integration module includes multiple domain-specific predictors, each domain-specific predictor integrates multiple prediction heads and respectively predicts the remaining service life results of the multivariate fusion features of the corresponding specific source domain samples, and fuses the remaining service life results of multiple prediction heads through an integrated learning strategy to generate a remaining service life prediction output of the gearbox gear;

[0017] S4, design a multi-level and multi-source domain adaptation strategy to perform local and global alignment of source and target domain feature distributions and output boundary alignment of the predictor, reduce the distribution differences between data features in different domains, and improve the domain-invariant representation capability of features;

[0018] S5, constructs the target network function using multi-level and multi-source domain adaptive losses and prediction losses of the source domain and target domain;

[0019] The target network function is optimized using back propagation and gradient descent algorithms to obtain the target network.

[0020] Constructing a life prediction model is convenient for use and can accurately predict the remaining service life of gears in the gearbox of the opposing grinding mill.

[0021] Further, in step S1, the method used to select the statistical features with the best trend and monotonicity in the gear speed signal of the main test gearbox and the box vibration acceleration signal is linear regression analysis, specifically:

[0022] y i (t) = β i,0 +β i,1 t,

[0023]

[0024] Among them, β i,0 , β i,1 and i (t) are the intercept, slope and statistical characteristic value corresponding to the i-th characteristic trend at time point t; is the average time, represents the average value of the i-th characteristic trend; K represents the length of statistical characteristic data.

[0025] Obtain the best statistical features of trend and monotonicity in the gear speed signal of the main test gearbox and the box vibration acceleration signal, which is beneficial for subsequent use.

[0026] Furthermore, the health indicator with the best trend and monotonicity selected in S1 is the pulse factor in the speed signal, which is:

[0027]

[0028] Among them, s f represents the impulse factor, x i represents the i-th sample point, n represents the number of samples, Represents the absolute value of the sample data mean.

[0029] The operation is simple and easy to use.

[0030] Furthermore, the final output of the multivariate fusion feature extractor in step S2 is:

[0031]

[0032] Among them, X n represents the nth input sample, F resnet (·)and Respectively represent the residual network ResNet branch and its corresponding output, F bigru (·)and They represent the bidirectional gated recurrent unit BIGRU branches and their corresponding outputs respectively; f nRepresents the final output of the multivariate fusion feature extractor.

[0033] Obtain fusion features for subsequent prediction.

[0034] Furthermore, the predictor in step S3 is:

[0035]

[0036] Where H represents the number of prediction heads in each predictor ensemble, represents the prediction result of the domain-specific predictor on the sample, S n，j Represents the remaining service life prediction result of the gearbox gears.

[0037] The predictor has a simple structure and can generate the desired prediction results.

[0038] Furthermore, the target network function is:

[0039] L=L y +k p ·L p +k w ·L w +k d ·L d ,

[0040] Among them, L represents the overall network optimization objective function, L y represents the total loss of remaining useful life prediction, k p , k w and k d Respectively represent L P , L W and L d The trade-off parameter of P represents the alignment loss at the predictor level, L W represents the Wasserstein distance loss, L d represents the loss of the domain discriminator in adversarial training.

[0041] Setting the target network function helps optimize the prediction results.

[0042] Furthermore, based on the model-free adaptive control MFAC theory and the introduction of pseudo partial derivative PPD and compact dynamic linearization CFDL model, the optimal control torque is calculated. The specific steps are as follows:

[0043] For a noise-free ideal single-input single-output discrete-time nonlinear system, it can be expressed as:

[0044] h(k+1)=f(h(k),...,h(kn i ),l(k),...,l(kn j))

[0045] Wherein, h(k) represents the input of the system at time k, i.e., the remaining service life result of the gear predicted by the remaining service life prediction model based on the monitoring data; l(k) represents the output of the system at time k, i.e., the load of the gear during operation; n i , n j are two unknown positive numbers, f(·) is an unknown nonlinear function;

[0046] When the function f(·) is continuously differentiable and the system satisfies the Lipschitz condition, that is, for any k1≠k2, k1,k2≥0 and

[0047] l(k1)≠l(k2) then:

[0048] |h(k1+1)-h(k2+1)|≤b||l(k1)-l(k2)||

[0049] Where b is a constant greater than 0;

[0050] For nonlinear systems that meet the above conditions, they can be converted into CFDL models by introducing a time-varying parameter called PPD, namely:

[0051] h(k+1)=h(k)+φ(k)Δl(k)

[0052] Where φ(k) is the PPD time-varying parameter;

[0053] The MFAC control input criterion function introduces a λ weighting factor to limit the change range of the control input while maintaining the expected value of the system output in order to avoid its negative impact or over-adjustment on system stability when predicting and adjusting the control input. It can be expressed as:

[0054] J(l(k))=|h * (k+1)-h(k+1)| 2 +λ|l(k)-l(k-1)| 2

[0055] Where, J(l(k)) represents the control input criterion function;

[0056] Substituting the linear model of the system into the criterion function and taking the derivative of the control input l(k), setting the derivative equal to zero and introducing the step size factor ρ, the final control algorithm can be obtained, namely:

[0057]

[0058] Among them, l(k) represents the output load at time k, ρ represents the step size factor, λ represents the weight factor, φ c(k) represents the time-varying parameter of PPD at time k, h * (k+1) represents the expected remaining life value of the system input at time k+1, and h(k) represents the remaining life prediction value output by the life prediction model at time k.

[0059] The PPD estimation algorithm and algorithm reset mechanism are expressed as:

[0060]

[0061] in, represents the PPD of φ at time k c The estimated value of (k), yes , η represents the introduced step size factor, μ represents the penalty factor for the PPD estimate, Δl(k) represents the output torque difference between time k and time k-1, Δh(k) represents the remaining service life prediction difference between time k and time k-1, and ε is a sufficiently small difference.

[0062] Based on MFAC theory, pseudo partial derivative (PPD) and compact form dynamic linearization (CFDL) model are introduced to realize dynamic adjustment of control parameters. At the same time, weighted criterion function, step size factor and algorithm reset mechanism are added to the control algorithm to enhance the robustness and flexibility of the control algorithm.

[0063] Furthermore, based on the optimal control condition parameters, the method for controlling the operating condition of the gearbox gear of the vertical mill is as follows:

[0064] When the difference between the predicted remaining life of the vertical mill gearbox gear and the current expected remaining life exceeds 10 hours, the counting variables a and b are increased according to the direction of the deviation:

[0065] If the predicted life of the gear of the vertical mill gearbox exceeds the current expected remaining life by more than 10 hours, the value of the counting variable a increases by 1; conversely, if the predicted life of the gear of the vertical mill gearbox is less than the current expected remaining life by more than 10 hours, the value of the counting variable b increases by 1;

[0066] If the accumulated value of the counting variable a or b reaches or exceeds 2, the operating condition of the vertical mill gearbox gear is adjusted based on the optimal control condition parameters until the absolute error between the predicted life of the vertical mill gearbox gear and the current expected remaining life is reduced to within 10 hours. At this time, the control operation is stopped and the counts a and b are reset to zero.

[0067] Extending the actual service life of the vertical mill gearbox gears to close to the expected service life ensures the working efficiency and long-term stability of the gear transmission system and reduces the economic losses and safety risks caused by accidental damage to the gears.

[0068] The present invention also provides an online intelligent control system for gear life of a vertical mill gearbox based on the method of the present invention, comprising a data acquisition module, a LabVIEW host computer unit and an algorithm engine module;

[0069] The data acquisition module includes an encoder and a vibration sensor, which are used to respectively collect the gear speed signal of the main test gearbox and the vibration acceleration signal of the box body;

[0070] The LabVIEW host computer unit is used to receive the collected gear speed signal and box vibration acceleration signal of the main test gear box, and determine whether the gear has entered the degradation stage;

[0071] The algorithm engine module is connected to the LabVIEW host computer unit. The algorithm engine module is used to predict the remaining life of the gears in the vertical mill gearbox and transmit it back to the LabVIEW host computer unit. The LabVIEW host computer unit is provided with a model-free adaptive control MFAC module. The model-free adaptive control MFAC module outputs optimal control operating condition parameters based on the remaining life of the gears in the vertical mill gearbox and the current expected remaining life. The LabVIEW host computer unit performs gear operating condition control based on the optimal control operating condition parameters.

[0072] This system uses various unit modules to achieve accurate prediction and online intelligent control of the remaining service life of the vertical mill gearbox gears, successfully extending the actual service life of the vertical mill gearbox gears to close to the expected service life, ensuring the working efficiency and long-term stability of the gear transmission system, and reducing the economic losses and safety risks caused by accidental damage to the gears. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a structural schematic diagram of the online intelligent control system for the gear life of a vertical mill gearbox of the present invention;

[0074] Figure 2 It is a schematic flow chart of constructing a life prediction model of the online intelligent control method for the life of a vertical mill gearbox gear of the present invention;

[0075] Figure 3 It is a schematic flow chart of the operating conditions of the gears in the vertical mill gearbox of the online intelligent control method for the life of the gears in the vertical mill gearbox of the present invention. DETAILED DESCRIPTION

[0076] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0077] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0078] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0079] The present invention discloses an online intelligent control method for the gear life of a vertical mill gearbox, which can optimize the gear operating conditions online and intelligently, effectively extend the actual service life of the gears in the vertical mill gearbox to make them closer to the expected service life, and provide guarantee for the long-term and efficient operation of the equipment. The online intelligent control method for the gear life of a vertical mill gearbox comprises the following steps:

[0080] Collect the gear speed signal and the vibration acceleration signal of the main test gearbox for 1s (interval) every 5 minutes;

[0081] Based on the 3σ criterion, determine whether the gear has entered the degradation stage. If the gear has entered the degradation stage, build a life prediction model and use the life prediction model to predict the remaining life of the vertical mill gearbox gear;

[0082] Combining the remaining life prediction results of the vertical mill gearbox gear and the current expected remaining life, the optimal control torque is calculated based on the model-free adaptive control MFAC theory and the introduction of pseudo partial derivative PPD and compact dynamic linearization CFDL model;

[0083] Based on the optimal control operating condition parameters, the operating conditions of the vertical mill gearbox gears are adjusted.

[0084] In a preferred embodiment of the present invention, Figure 2 As shown in Figure 2, the construction method of the life prediction model is:

[0085] S1, select the statistical features with the best trend and monotonicity of the gear speed signal and the vibration acceleration signal of the main gearbox in the whole life experiment as the health index of the gear, form a health index sequence in the whole life experiment, and construct the life sample on the health index sequence by sliding window method with a window length of 512 and a step size of 1;

[0086] S2, constructing a multivariate fusion feature extractor, extracting multivariate feature information related to degradation state and remaining service life in the life samples in parallel, fusing the multivariate feature information, and generating a final output of the multivariate fusion feature extractor;

[0087] S3, constructing a prediction integration module, the prediction integration module includes multiple domain-specific predictors, each domain-specific predictor integrates multiple prediction heads and respectively predicts the remaining service life results of the multivariate fusion features of the corresponding specific source domain samples, and fuses the remaining service life results of multiple prediction heads through an integrated learning strategy to generate a remaining service life prediction output of the gearbox gear;

[0088] S4, design a multi-level and multi-source domain adaptation strategy to perform local and global alignment of source and target domain feature distributions and output boundary alignment of the predictor, reduce the distribution differences between data features in different domains, and improve the domain-invariant representation capability of features;

[0089] S5, constructs the target network function using multi-level and multi-source domain adaptive losses and prediction losses of the source domain and target domain;

[0090] The target network function is optimized using back propagation and gradient descent algorithms to obtain the target network.

[0091] In a preferred embodiment of the present invention, the method used in step S1 to select the statistical features with the best trend and monotonicity in the gear speed signal of the main test gearbox and the box vibration acceleration signal is linear regression analysis, specifically:

[0092] y i (t) = β i,0 +β i,1 t,

[0093]

[0094] Among them, βi,0, βi,1 and y i (t) are the intercept, slope and statistical characteristic value corresponding to the i-th characteristic trend at time point t; is the average time, represents the average value of the i-th characteristic trend; K represents the length of statistical characteristic data.

[0095] Preferably, the health indicator with the best trend and monotonicity selected in S1 is the pulse factor in the speed signal, which is:

[0096]

[0097] Among them, s f represents the impulse factor, x irepresents the i-th sample point, n represents the number of samples, Represents the absolute value of the sample data mean.

[0098] In a preferred embodiment of the present invention, the final output of the multivariate fusion feature extractor in step S2 is:

[0099]

[0100] Among them, X n represents the nth input sample, F resnet (·)and Respectively represent the residual network ResNet branch and its corresponding output, F bigru (·)and They represent the bidirectional gated recurrent unit (BIGRU) branches and their corresponding outputs respectively; fn represents the final output of the multi-fusion feature extractor.

[0101] More preferably, the predictor in step S3 is:

[0102]

[0103] Where H represents the number of prediction heads in each predictor ensemble, represents the prediction result of the domain-specific predictor for the sample, Sn, j Represents the remaining service life prediction result of the gearbox gears.

[0104] In a preferred embodiment of the present invention, the target network function is:

[0105] L=L y +k p ·L p +k w ·L w +k d ·L d ,

[0106] Among them, L represents the overall network optimization objective function, L y represents the total loss of remaining useful life prediction, k p , k w and k d Respectively represent L P , L W and L d The trade-off parameter of P represents the alignment loss at the predictor level, L W represents the Wasserstein distance loss, L d represents the loss of the domain discriminator in adversarial training.

[0107] In a preferred embodiment of the present invention, based on the model-free adaptive control MFAC theory and introducing the pseudo partial derivative PPD and the compact dynamic linearization CFDL model, the control parameters are dynamically adjusted and the optimal control torque is calculated. The specific steps are:

[0108] For a noise-free ideal single-input single-output discrete-time nonlinear system, it can be expressed as:

[0109] h(k+1)=f(h(k),...,h(kn i ),l(k),...,l(kn j ))

[0110] Wherein, h(k) represents the input of the system at time k, i.e., the remaining service life result of the gear predicted by the remaining service life prediction model based on the monitoring data; l(k) represents the output of the system at time k, i.e., the load of the gear during operation; n i , n j are two unknown positive numbers, f(·) is an unknown nonlinear function;

[0111] When the function f(·) is continuously differentiable and the system satisfies the Lipschitz condition, that is, for any k1≠k2, k1,k2≥0 and

[0112] l(k1)≠l(k2) then:

[0113] |h(k1+1)-h(k2+1)|≤b||l(k1)-l(k2)||

[0114] Where b is a constant greater than 0;

[0115] For nonlinear systems that meet the above conditions, they can be converted into CFDL models by introducing a time-varying parameter called PPD, namely:

[0116] h(k+1)=h(k)+φ(k)Δl(k)

[0117] Where φ(k) is the PPD time-varying parameter;

[0118] The MFAC control input criterion function introduces a λ weighting factor to limit the change range of the control input while maintaining the expected value of the system output in order to avoid its negative impact or over-adjustment on system stability when predicting and adjusting the control input; it can be expressed as:

[0119] J(l(k))=|h * (k+1)-h(k+1)| 2 +λ|l(k)-l(k-1)| 2

[0120] Where, J(l(k)) represents the control input criterion function;

[0121] Substituting the linear model of the system into the criterion function and taking the derivative of the control input l(k), setting the derivative equal to zero and introducing the step size factor ρ, the final control algorithm can be obtained, namely:

[0122]

[0123] Among them, l(k) represents the output load at time k, ρ represents the step size factor, λ represents the weight factor, φ c (k) represents the time-varying parameter of PPD at time k, h * (k+1) represents the expected remaining life value of the system input at time k+1, and h(k) represents the remaining life prediction value PPD output by the life prediction model at time k. The estimation algorithm and algorithm reset mechanism are expressed as:

[0124]

[0125] in, represents the PPD of φ at time k c The estimated value of (k), yes , η represents the introduced step size factor, μ represents the penalty factor for the PPD estimate, Δl(k) represents the output torque difference between time k and time k-1, Δh(k) represents the remaining service life prediction difference between time k and time k-1, and ε is a sufficiently small difference.

[0126] In a preferred embodiment of the present invention, Figure 3 As shown, based on the optimal control condition parameters, the method for controlling the operating condition of the vertical mill gearbox gear is as follows:

[0127] When the difference between the predicted remaining life of the vertical mill gearbox gear and the current expected remaining life exceeds 10 hours, the counting variables a and b are increased according to the direction of the deviation:

[0128] If the predicted life of the gear of the vertical mill gearbox exceeds the current expected remaining life by more than 10 hours, the value of the counting variable a increases by 1; conversely, if the predicted life of the gear of the vertical mill gearbox is less than the current expected remaining life by more than 10 hours, the value of the counting variable b increases by 1;

[0129] If the accumulated value of the counting variable a or b reaches or exceeds 2, the operating condition of the vertical mill gearbox gear is adjusted based on the optimal control condition parameters until the absolute error between the predicted life of the vertical mill gearbox gear and the current expected remaining life is reduced to within 10 hours. At this time, the control operation is stopped and the counts a and b are reset to zero.

[0130] The present invention also provides an online intelligent control system for the gear life of a vertical mill gearbox based on the method of the present invention, such as Figure 1 As shown, it includes a data acquisition module, a LabVIEW host computer unit and an algorithm engine module. The vertical mill gearbox includes a drive motor, a main test gearbox, a companion test gearbox and a load motor. A control cabinet (such as a Siemens S7-1200 PLC) is set up to control the drive motor and the load motor. A thin oil station is set up to lubricate the main test gearbox and the companion test gearbox through two independent oil circuits.

[0131] The data acquisition module includes an encoder and a vibration sensor, which are used to respectively collect the gear speed signal of the main test gearbox and the box vibration acceleration signal.

[0132] The LabVIEW host computer unit works based on the DAQmx driver data acquisition module provided by NI, receives and stores the raw data from the encoder and vibration sensor in real time, and monitors whether the gear enters the degradation stage in real time based on the 3σ criterion.

[0133] After the LabVIEW host computer receives and stores the raw data in real time through the acquisition card, it packages it and sends it to the algorithm engine module deployed on the local server. If degradation is detected, the LabVIEW host computer unit will package the stored data into JSON format and send it to the algorithm engine module deployed on the local server through the HTTP communication protocol.

[0134] The algorithm engine module is a self-developed algorithm engine module that can run on a local server. This module allows the embedding of Python-based algorithm models and provides multiple input and output interfaces to achieve real-time communication with a variety of data acquisition software.

[0135] The algorithm engine module is electrically connected to the LabVIEW host computer unit. The algorithm engine module is used to predict the remaining life of the gears in the vertical mill gearbox and transmit it back to the LabVIEW host computer unit based on the HTTP communication protocol. The LabVIEW host computer unit is provided with a model-free adaptive control MFAC module. The model-free adaptive control MFAC module uses MATLABscript nodes in LabVIEW to output the optimal control operating condition parameters based on the remaining life of the gears in the vertical mill gearbox and the current expected remaining life. The LabVIEW host computer unit performs gear operating condition control based on the optimal control operating condition parameters.

[0136] After receiving the remaining service life results and expected remaining service life output from the algorithm engine, the MFAC module outputs the optimal operating parameters of the vertical mill gearbox gear under the current actual remaining service life through the control algorithm. In addition, considering the stability of the life prediction algorithm, the control algorithm is further optimized by introducing a secondary judgment mechanism to ensure the stability of the system.

[0137] After the LabVIEW host computer unit obtains the control operating parameters output by the MFAC module, it communicates with the programmable logic controller (PLC) module in the control cabinet through the Ethernet cable to realize real-time control of the gearbox gear operating status.

[0138] This system uses various unit modules to achieve accurate prediction and online intelligent control of the remaining service life of the vertical mill gearbox gears, successfully extending the actual service life of the vertical mill gearbox gears to close to the expected service life, ensuring the working efficiency and long-term stability of the gear transmission system, and reducing the economic losses and safety risks caused by accidental damage to the gears.

[0139] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0140] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for online intelligent control of gear life of a vertical mill gearbox, characterized in that: The steps include: Intermittently collect the gear speed signal of the main test gearbox and the box vibration acceleration signal; Based on the 3σ criterion, determine whether the gear has entered the degradation stage. If the gear has entered the degradation stage, build a life prediction model and use the life prediction model to predict the remaining life of the vertical mill gearbox gear; Combining the remaining life prediction results of the vertical mill gearbox gear and the current expected remaining life, the optimal control torque is calculated based on the model-free adaptive control MFAC theory and the introduction of pseudo partial derivative PPD and compact dynamic linearization CFDL model; Based on the optimal control operating condition parameters, the operating conditions of the vertical mill gearbox gears are adjusted.

2. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 1, characterized in that: The method for constructing the life prediction model is: S1, select the statistical features with the best trend and monotonicity of the gear speed signal and the vibration acceleration signal of the main gearbox in the whole life experiment as the health index of the gear, form a health index sequence in the whole life experiment, and construct the life sample on the health index sequence by sliding window method with a window length of 512 and a step size of 1; S2, constructing a multivariate fusion feature extractor, extracting multivariate feature information related to degradation state and remaining service life in the life samples in parallel, fusing the multivariate feature information, and generating a final output of the multivariate fusion feature extractor; S3, constructing a prediction integration module, the prediction integration module includes multiple domain-specific predictors, each domain-specific predictor integrates multiple prediction heads and respectively predicts the remaining service life results of the multivariate fusion features of the corresponding specific source domain samples, and fuses the remaining service life results of multiple prediction heads through an integrated learning strategy to generate a remaining service life prediction output of the gearbox gear; S4, design a multi-level and multi-source domain adaptation strategy to perform local and global alignment of source and target domain feature distributions and output boundary alignment of the predictor, reduce the distribution differences between data features in different domains, and improve the domain-invariant representation capability of features; S5, constructs the target network function using multi-level and multi-source domain adaptive losses and prediction losses of the source domain and target domain; The target network function is optimized using back propagation and gradient descent algorithms to obtain the target network.

3. The online intelligent control method for vertical mill gearbox gear life according to claim 2, characterized in that: In step S1, the method used to select the statistical features with the best trend and monotonicity in the gear speed signal of the main test gearbox and the box vibration acceleration signal is linear regression analysis, specifically: y i (t)=β i,0 +b i,1 t, Among them, β i,0 , β i,1 and i (t) are the intercept, slope and statistical characteristic value corresponding to the i-th characteristic trend at time point t; is the average time, represents the average value of the i-th characteristic trend; K represents the length of statistical characteristic data.

4. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 2, characterized in that: The health indicator with the best trend and monotonicity selected in S1 is the pulse factor in the speed signal, which is: Among them, s f represents the impulse factor, x i represents the i-th sample point, n represents the number of samples, Represents the absolute value of the sample data mean.

5. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 2, characterized in that: The final output of the multivariate fusion feature extractor in step S2 is: Among them, X n represents the nth input sample, F resnet (·)and Respectively represent the residual network ResNet branch and its corresponding output, F bigru (·)and They represent the bidirectional gated recurrent unit BIGRU branches and their corresponding outputs respectively; f n Represents the final output of the multivariate fusion feature extractor.

6. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 2, characterized in that: The predictor in step S3 is: Where H represents the number of prediction heads in each predictor ensemble, represents the prediction result of the domain-specific predictor on the sample, S n,j Represents the remaining service life prediction result of the gearbox gears.

7. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 2, characterized in that: The target network function is: L=L y +k p ·L p +k w ·L w +k d ·L d , Among them, L represents the overall network optimization objective function, L y represents the total loss of remaining useful life prediction, k p , k w and k d Respectively represent L P , L W and L d The trade-off parameter of P represents the alignment loss at the predictor level, L W represents the Wasserstein distance loss, L d represents the loss of the domain discriminator in adversarial training.

8. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 1, characterized in that: Based on the model-free adaptive control MFAC theory and the introduction of pseudo partial derivative PPD and compact dynamic linearization CFDL model, the optimal control torque is calculated. The specific steps are as follows: For a noise-free ideal single-input single-output discrete-time nonlinear system, it can be expressed as: h(k+1)=f(h(k),...,h(k-n i ),l(k),...,l(k-n j )) Wherein, h(k) represents the input of the system at time k, i.e., the remaining service life result of the gear predicted by the remaining service life prediction model based on the monitoring data; l(k) represents the output of the system at time k, i.e., the load of the gear during operation; n i , n j are two unknown positive numbers, f(·) is an unknown nonlinear function; When the function f(·) is continuously differentiable and the system satisfies the Lipschitz condition, that is, for any k1≠k2, k1,k2≥0 and l(k1)≠l(k2) then: |h(k1+1)-h(k2+1)|≤b||l(k1)-l(k2)|| Where b is a constant greater than 0; For nonlinear systems that meet the conditions, they can be converted into CFDL models by introducing a time-varying parameter called PPD, namely: h(k+1)=h(k)+φ(k)Δl(k) Where φ(k) is the PPD time-varying parameter; The MFAC control input criterion function introduces a λ weighting factor to limit the change range of the control input while maintaining the expected value of the system output in order to avoid its negative impact or over-adjustment on system stability when predicting and adjusting the control input. It is expressed as: J(l(k))=|h * (k+1)-h(k+1)| 2 +λ|l(k)-l(k-1)| 2 Where, J(l(k)) represents the control input criterion function; Substitute the linear model of the system into the criterion function and derive the control input l(k), set the derivative equal to zero and introduce the step size factor ρ to obtain the final control algorithm, namely: Among them, l(k) represents the output load at time k, ρ represents the step size factor, λ represents the weight factor, φ c (k) represents the time-varying parameter of PPD at time k, h * (k+1) represents the expected remaining life value of the system input at time k+1, and h(k) represents the remaining life prediction value output by the life prediction model at time k; The PPD estimation algorithm and algorithm reset mechanism are expressed as: in, represents the PPD of φ at time k c The estimated value of (k), yes , η represents the introduced step size factor, μ represents the penalty factor for the PPD estimate, Δl(k) represents the output torque difference between time k and time k-1, Δh(k) represents the remaining service life prediction difference between time k and time k-1, and ε is a sufficiently small difference.

9. The method for online intelligent control of gear life of a vertical mill gearbox as claimed in claim 1, characterized in that: Based on the optimal control condition parameters, the method for adjusting the operating condition of the vertical mill gearbox gear is as follows: When the difference between the predicted remaining life of the vertical mill gearbox gear and the current expected remaining life exceeds 10 hours, the counting variables a and b are increased according to the direction of the deviation: If the predicted life of the gear of the vertical mill gearbox exceeds the current expected remaining life by more than 10 hours, the value of the counting variable a increases by 1; conversely, if the predicted life of the gear of the vertical mill gearbox is less than the current expected remaining life by more than 10 hours, the value of the counting variable b increases by 1; If the accumulated value of the counting variable a or b reaches or exceeds 2, the operating condition of the vertical mill gearbox gear is adjusted based on the optimal control condition parameters until the absolute error between the predicted life of the vertical mill gearbox gear and the current expected remaining life is reduced to within 10 hours. At this time, the control operation is stopped and the counts a and b are reset to zero.

10. An online intelligent control system for gear life of a vertical mill gearbox based on the method according to any one of claims 1 to 9, characterized in that: It includes data acquisition module, LabVIEW host computer unit and algorithm engine module; The data acquisition module includes an encoder and a vibration sensor, which are used to respectively collect the gear speed signal of the main test gearbox and the vibration acceleration signal of the box body; The LabVIEW host computer unit is used to receive the collected gear speed signal and box vibration acceleration signal of the main test gear box, and determine whether the gear has entered the degradation stage; The algorithm engine module is connected to the LabVIEW host computer unit. The algorithm engine module is used to predict the remaining life of the gears in the vertical mill gearbox and transmit it back to the LabVIEW host computer unit. The LabVIEW host computer unit is provided with a model-free adaptive control MFAC module. The model-free adaptive control MFAC module outputs optimal control operating condition parameters based on the remaining life of the gears in the vertical mill gearbox and the current expected remaining life. The LabVIEW host computer unit performs gear operating condition control based on the optimal control operating condition parameters.

Images