Mechanism-data hybrid model construction method for online prediction of cutting moment of machine tool spindle

Abstract

The invention relates to the technical field of numerical control machine tools, in particular to a mechanism-data hybrid model construction method for online prediction of cutting torque of a machine tool spindle. The method comprises the steps that main shaft three-phase current, main shaft rotating speed and main shaft angular acceleration serve as input, main shaft cutting torque serves as output, a recurrent neural network model is constructed to serve as a basic model, a motor torque estimation model serves as input of the basic model, and therefore a mechanism-data hybrid model is constructed; wherein the recurrent neural network model is a gating recurrent unit neural network model or a long short-term memory neural network model; according to the method, the training efficiency and generalization ability of the model are effectively improved, and the accuracy and reliability of online prediction in the application process of online prediction of the cutting torque of the machine tool spindle are ensured.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool technology, and specifically to a mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque. Background Technology

[0002] CNC machining is a core component of modern manufacturing and is widely used in high-end manufacturing fields such as aerospace, rail transportation, and shipbuilding. With the increasing performance requirements of equipment, the characteristics of high precision requirements for parts, difficult material processing, large size, and long processing cycles are becoming increasingly prominent. Cutting force is the most direct and effective monitorable physical quantity reflecting the cutting state. Cutting force monitoring can be applied to various scenarios such as tool wear / breakage monitoring, chatter identification, and adaptive machining, which is crucial for ensuring machining quality, ensuring equipment safety, and improving machining efficiency.

[0003] Traditional cutting force monitoring methods rely on specialized measuring equipment such as benchtop force gauges and rotary force gauges, which are difficult to apply to actual production processes in terms of practicality, reliability, and economy. Cutting forces can be predicted online using machine tool servo monitoring signals such as motor output torque, speed, and acceleration. Since this method does not intrude on the machine tool workspace and is relatively low in cost, it has the potential to be applied in actual production processes. Therefore, in order to make full use of motor output torque, speed, and acceleration in neural network models, there is an urgent need for a mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque.

[0004] For example, a Chinese patent, publication number CN117733646A, publication date March 22, 2024, entitled "A Data-Driven Method for Monitoring Machine Tool Spindle Cutting Torque," describes a data-driven method for monitoring machine tool spindle cutting torque. This method synchronously monitors servo signals such as the output torque or three-phase current of the machine tool spindle motor, spindle motion status, and spindle cutting torque. Cutting experiments are conducted to obtain a spindle servo monitoring signal-cutting torque dataset. This dataset is then trained using a sequence-sequence supervised training method to create a data-driven time-series single-step prediction model. In practical applications, the corresponding spindle servo monitoring signals are input into the prediction model for single-step real-time prediction or sequence-sequence prediction, thereby achieving online monitoring of the spindle cutting torque.

[0005] Compared with traditional force sensing elements, the above-mentioned patents have significantly reduced costs and do not affect the machine tool workspace. However, the above-mentioned patents are purely data-driven models with poor interpretability and rely on the assumption of independent and identically distributed models, so the generalization ability of the models is relatively poor. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque, which can improve the training efficiency and generalization ability of the model and ensure the accuracy and reliability of online prediction of spindle cutting torque during application.

[0007] To achieve the above-mentioned technical effects, the technical solution of this application is as follows:

[0008] The mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque uses the three-phase current, spindle speed, and spindle angular acceleration of the spindle as inputs and the spindle cutting torque as output to construct a recurrent neural network model as the basic model. Then, the motor torque estimation model is used as the input of the basic model to construct the mechanism-data hybrid model. Among them, the recurrent neural network model is a gated recurrent unit neural network model or a long short-term memory neural network model.

[0009] Furthermore, the motor output torque is obtained through the motor torque estimation model, and the motor output torque is used as the input to the mechanism-data hybrid model. The specific solution steps for the motor output torque are as follows:

[0010] S1: The stator excitation I is obtained by transforming the three-phase current of the main shaft using the motor torque estimation model. d and stator torque current I q ;

[0011] S2: Stator excitation I obtained from S1 d Find the rotor flux linkage ψ r And based on rotor flux ψ r and stator torque current I q The calculated speed difference ω sl Finally, the obtained speed difference ω sl The rotor flux linkage angle θ is obtained;

[0012] S3: Based on the stator torque current I obtained in step S2 q and rotor flux ψ r The output torque of the motor is calculated and used as the input to the mechanism-data hybrid model.

[0013] Furthermore, the motor torque estimation model is used as the input to the gated recurrent unit neural network model, as shown in the following formula:

[0014]

[0015] In the formula, T m Represents the spindle cutting torque, GRU represents the gated loop unit neural network model, and I represents the three-phase spindle current [I u ,I v,I w ], ω represents the spindle speed, The model represents the spindle angular acceleration, h represents the hidden state of the model, t represents the current time, and t-1 represents the previous time; G represents the motor torque estimation model, which takes the three-phase current of the spindle and the spindle speed as inputs. t (I,ω) represents the motor output torque value at the current moment.

[0016] Furthermore, the motor torque estimation model is used as input to the long short-term memory neural network model, as shown in the following formula:

[0017]

[0018] In the formula, T m The spindle cutting torque is represented by LSTM, the long short-term memory neural network model is represented by I, and the three-phase current of the spindle is represented by [I]. u ,I v ,I w ], ω represents the spindle speed, Let represent the spindle angular acceleration, h represent the hidden state of the model, c represent the element state of the model, t represent the current time, t-1 represent the previous time, and G represent the motor torque estimation model, which takes the three-phase current and speed of the spindle as inputs. t (I,ω) represents the motor output torque value at the current moment.

[0019] Furthermore, the motor torque estimation model adopts the motor torque estimation model of the current-type rotor flux observer.

[0020] Furthermore, step S1 is specifically implemented as follows: The motor torque estimation model of the current-type rotor flux observer is used to estimate the three-phase current I. u I v and I w The stator excitation I is obtained by performing 3s / 2s and 2s / 2r transformations. d and stator torque current I q Its transformation formula is:

[0021]

[0022] In the formula, I d I represents the excitation of the stator. q C represents the stator torque current. 2s / 2r Represents the 2s / 2r transformation, C 3s / 2s This represents the 3s / 2s transformation, and θ represents the rotor flux linkage angle.

[0023] Furthermore, based on the stator excitation I d and stator torque current I qFind the rotor flux linkage ψ r The specific formula is:

[0024]

[0025] In the formula, τ r L is the rotor time constant of the motor. m Let be the stator inductance, and s represent the differential operator.

[0026] Furthermore, based on the rotor flux ψ r and stator torque current I q The speed difference ω was calculated. sl The specific formula is:

[0027]

[0028] In the formula, τ r L is the rotor time constant of the motor. m For stator inductance, τ r Let ψ be the rotor time constant of the motor, and ψ be the rotor flux linkage. r .

[0029] Furthermore, through the speed difference ω sl The specific formula for obtaining the rotor flux linkage angle θ is:

[0030]

[0031] In the formula, ω r The expression represents the product of the actual speed of the motor and the number of pole pairs of the motor, and s represents the differential operator.

[0032] Furthermore, the product ω of the motor's actual speed and the number of pole pairs is... r The specific calculation formula is as follows:

[0033] ω r =ωn p ;

[0034] In the formula, ω represents the actual speed of the motor, and n p This indicates the number of pole pairs of the motor.

[0035] Furthermore, the motor output torque finally obtained in step S3 is specifically calculated using the following formula:

[0036]

[0037] In the formula, T e n represents the output torque of the motor. p I represents the number of pole pairs of the motor. q Represents torque current, ψ r L represents the rotor flux linkage. rL represents the rotor inductance. m It is the stator inductance.

[0038] Furthermore, when it is necessary to obtain the stator torque current I at the current moment... q,t and the rotor flux linkage ψ of the stator r,t The specific formula for calculating the motor output torque at the current moment is as follows:

[0039]

[0040] In the formula, G t (I,ω) and T e,t n represents the current motor output torque value. p I represents the number of pole pairs of the motor. q,t ψ represents the torque current at the current moment. r,t L represents the rotor flux linkage at the current moment. r L represents the rotor inductance. m It is the stator inductance.

[0041] Furthermore, the inputs to the mechanism-data hybrid model in step S3 include the motor output torque, the three-phase current of the spindle, the spindle speed, and the spindle angular acceleration; the output of the mechanism-data hybrid model is the spindle cutting torque.

[0042] Based on the above technical solution, the beneficial effects of the present invention are as follows:

[0043] 1. The method of the present invention utilizes the mechanism characteristics of a three-phase asynchronous motor and the principle of spindle cutting torque prediction to first convert a portion of the input into an essential physical quantity that is more relevant to the spindle cutting torque. This is equivalent to adding a beneficial feature transformation to the recurrent neural network model. Compared with the prior art, which directly uses the recurrent neural network model to predict the spindle cutting torque, the present invention effectively improves the training efficiency and generalization ability of the model.

[0044] 2. The method of the present invention reduces the dependence on precise motor parameters by adopting a motor torque estimation model with a current-type rotor flux observer, and has good real-time current signal performance and strong noise suppression capability.

[0045] 3. The method of the present invention uses the motor torque estimation model as the input of the basic model. Compared with the existing online prediction method for machine tool cutting force, the method of the present invention ensures the accuracy and reliability of online prediction in the application of online prediction of machine tool spindle cutting torque. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of a current-type rotor flux linkage observer.

[0047] Figure 2This is a schematic diagram of a milling experiment.

[0048] Figure 3 This is a schematic diagram comparing the model prediction performance of normal cutting tools in dataset 1.

[0049] Figure 4 This is a schematic diagram comparing the model prediction performance of the broken tools in dataset 1.

[0050] Figure 5 This is a schematic diagram comparing the model prediction performance of normal cutting tools in dataset 2.

[0051] Figure 6 This is a diagram showing the comparison of model prediction performance for damaged tools in dataset 2.

[0052] Figure 7 This is a schematic diagram comparing the model prediction performance of normal cutting tools in dataset 3.

[0053] Figure 8 This is a diagram showing the comparison of model prediction performance for damaged tools in dataset 3. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0055] Example 1

[0056] The mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque uses the three-phase current, spindle speed, and spindle angular acceleration of the spindle as inputs and the spindle cutting torque as output to construct a recurrent neural network model as the basic model. Then, the motor torque estimation model is used as the input of the basic model to construct the mechanism-data hybrid model. Among them, the recurrent neural network model is a gated recurrent unit neural network model or a long short-term memory neural network model.

[0057] Example 2

[0058] A mechanism-data hybrid model construction method for online prediction of machine tool spindle cutting torque is proposed. The method uses the three-phase spindle current, spindle speed, and spindle angular acceleration as inputs, and the spindle cutting torque as the output. A recurrent neural network model is constructed as the base model, and a motor torque estimation model is used as the input to the base model, thus constructing the mechanism-data hybrid model. The recurrent neural network model is either a gated recurrent unit neural network model or a long short-term memory neural network model. While gated recurrent unit neural network models or long short-term memory neural network models are mature existing technologies in this field, the innovation of this invention lies in using the motor torque estimation model as the input to the base model. The construction details of the gated recurrent unit neural network model or long short-term memory neural network model will not be elaborated here.

[0059] The motor output torque is obtained through a motor torque estimation model, and then used as the input to a mechanism-data hybrid model. The specific steps for solving the motor output torque are as follows:

[0060] S1: The stator excitation I is obtained by transforming the three-phase current of the main shaft using the motor torque estimation model. d and stator torque current I q ;

[0061] S2: Stator excitation I obtained from S1 d Find the rotor flux linkage ψ r And based on rotor flux ψ r and stator torque current I q The calculated speed difference ω sl Finally, the obtained speed difference ω sl The rotor flux linkage angle θ is obtained;

[0062] S3: Based on the stator torque current I obtained in step S2 q and rotor flux ψ r The output torque of the motor is calculated and used as the input to the mechanism-data hybrid model.

[0063] The motor torque estimation model is used as the input to the gated recurrent unit neural network model, and the specific formula is expressed as follows:

[0064]

[0065] In the formula, T m Represents the spindle cutting torque, GRU represents the gated loop unit neural network model, and I represents the three-phase spindle current [I u ,I v ,I w ], ω represents the spindle speed, The model represents the spindle angular acceleration, h represents the hidden state of the model, t represents the current time, and t-1 represents the previous time; G represents the motor torque estimation model, which takes the three-phase current of the spindle and the spindle speed as inputs. t (I,ω) represents the motor output torque value at the current moment.

[0066] The motor torque estimation model is used as input to the long short-term memory neural network model, and the specific formula is expressed as follows:

[0067]

[0068] In the formula, T m The spindle cutting torque is represented by LSTM, the long short-term memory neural network model is represented by I, and the three-phase current of the spindle is represented by [I]. u ,I v ,I w ], ω represents the spindle speed, Let represent the spindle angular acceleration, h represent the hidden state of the model, c represent the element state of the model, t represent the current time, t-1 represent the previous time, and G represent the motor torque estimation model, which takes the three-phase current and speed of the spindle as inputs. t (I,ω) represents the motor output torque value at the current moment.

[0069] Example 3

[0070] Based on Example 2, the motor torque estimation model adopts the motor torque estimation model of the current-type rotor flux observer.

[0071] like Figure 1 As shown, the specific method of step S1 is as follows: the motor torque estimation model of the current-type rotor flux observer is used to estimate the three-phase current I. u I v and I w The stator excitation I is obtained by performing 3s / 2s and 2s / 2r transformations. d and stator torque current I q Its transformation formula is:

[0072]

[0073] In the formula, I d I represents the excitation of the stator. q C represents the stator torque current. 2s / 2r Represents the 2s / 2r transformation, C 3s / 2s The 3s / 2s transformation represents the rotor flux linkage angle; where the 3s / 2s transformation refers to the transformation from 3-phase stationary coordinates to 2-phase stationary coordinates, and the 2s / 2r transformation refers to the transformation from 2-phase stationary coordinates to 2-phase rotating coordinates.

[0074] According to the stator excitation I d and stator torque current I q Find the rotor flux linkage ψ r The specific formula is:

[0075]

[0076] In the formula, τ r L is the rotor time constant of the motor. m Let be the stator inductance, and s represent the differential operator.

[0077] According to the rotor flux ψ r and stator torque current I q The speed difference ω was calculated. sl The specific formula is:

[0078]

[0079] In the formula, τ r L is the rotor time constant of the motor. m For stator inductance, τ r Let ψ be the rotor time constant of the motor, and ψ be the rotor flux linkage. r .

[0080] Through the speed difference ω sl The specific formula for obtaining the rotor flux linkage angle θ is:

[0081]

[0082] In the formula, ω r The expression represents the product of the actual speed of the motor and the number of pole pairs of the motor, and s represents the differential operator.

[0083] The product ω of the actual speed of the motor and the number of pole pairs of the motor. r The specific calculation formula is as follows:

[0084] ω r =ωn p ;

[0085] In the formula, ω represents the actual speed of the motor, and n p This indicates the number of pole pairs of the motor.

[0086] The final motor output torque obtained in step S3 is given by the following formula:

[0087]

[0088] In the formula, T e n represents the output torque of the motor. p I represents the number of pole pairs of the motor. q Represents torque current, ψr This indicates rotor flux linkage.

[0089] When it is necessary to obtain the stator torque current I at the current moment q,t and the rotor flux linkage ψ of the stator r,t The motor output torque at the current moment can be calculated using the following formula:

[0090]

[0091] In the formula, G t (I,ω) and T e,t n represents the current motor output torque value. p I represents the number of pole pairs of the motor. q,t ψ represents the torque current at the current moment. r,t L represents the rotor flux linkage at the current moment. r This is the rotor inductance.

[0092] The inputs to the mechanism-data hybrid model in step S3 include the motor output torque, the three-phase current of the spindle, the spindle speed, and the spindle angular acceleration; the output of the mechanism-data hybrid model is the spindle cutting torque.

[0093] Example 4

[0094] This implementation uses a torque estimation model based on a current-type rotor flux observer and a long short-term memory (LSTM) neural network to construct a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque. Since this embodiment uses an LSTM model, it is more intuitive to name the mechanism-data hybrid model the mechanism-LSTM neural network hybrid model in the following text.

[0095] In this implementation case, since the motor output torque estimation is relatively accurate, a separate three-phase current input I is no longer added to the input. t Therefore, the mechanism-LSTM neural network hybrid model in this implementation case can be expressed as:

[0096] T m,t =LSTM([G t (I,ω),ω t ],h t-1 ,c t-1 );

[0097] Here, LSTM represents a neural network containing one LSTM unit and one fully connected layer, G represents the motor torque estimation model based on the current-mode rotor flux observer at time t, and the model input is set to the output of the motor torque estimation model and the speed, i.e., x. t =[G t (I,ω),ωt ].

[0098] LSTM neural networks are a conventional existing technology in this field. An LSTM unit has three gate structures: a forget gate, an input gate, and an output gate.

[0099] The output f of the forget gate at time t t It can be represented as:

[0100] f t =σ(W if x t +b if +W hf h t-1 +b hf );

[0101] Where, x t Given the input at time t, h t-1 Let be the hidden state at time t-1, σ represent the sigmoid function, W be the corresponding weight matrix, and b be the corresponding bias vector.

[0102] The output i of the input gate at time t t It can be represented as:

[0103] i t =σ(W ii x t +b ii +W hi h t-1 +b hi );

[0104] To generate the cell state c at time t t First, the candidate cell state at time t is calculated.

[0105]

[0106] Furthermore, after obtaining the output f of the forget gate... t Input gate output i t and candidate cell status After that, the cell state c at time t t It can be represented as:

[0107]

[0108] Where ⊙ represents the Hadamard product, i.e., element-wise multiplication, and the output o of the output gate at time t. t It can be represented as:

[0109] o t =σ(W io x t +bio +W ho h t-1 +b ho );

[0110] Hidden state h of LSTM cell at time t t From cell state c t and output gate output o t The decision was made jointly, and the calculation formula is as follows:

[0111] h t =o t ⊙tanh(c t );

[0112] Ultimately, the mechanism—the LSTM neural network outputs the spindle cutting torque T at time t. m,t By h t After transformation by a linear fully connected layer, the following is obtained:

[0113] T m,t =W l h t +b l ;

[0114] The LSTM neural network model used for comparison is set as input to the master shaft three-phase current and speed, i.e., x. t =[I t ,ω t Therefore, it can be expressed as:

[0115] T m,t =LSTM([I t ,ω t ],h t-1 ,c t-1 );

[0116] The core component of the motor torque estimation model G is Figure 1 The current-mode rotor flux linkage observer shown. For three-phase current I... u I v and I w I is obtained by performing 3s / 2s and 2s / 2r transformations. d and I q The transformation formula is expressed as:

[0117]

[0118] Among them, I d and I q C represents the excitation and torque currents of the stator, respectively. 2s / 2r Represents the 2s / 2r transformation, C 3s / 2s This represents the 3s / 2s transformation, and θ represents the rotor flux linkage angle.

[0119] Based on I d and I q The rotor flux linkage ψ can be calculated. r and speed difference ω sl Then, the rotor flux linkage angle θ is obtained, and the calculation formula is:

[0120]

[0121] Where, τ r L is the rotor time constant of the motor. m For stator inductance, ω r The value represents the product of the actual speed of the motor and the number of pole pairs, and s represents the differential operator. In practical applications, the above current-type observer should be time-discrete, similar to the LSTM part of the discrete-time hybrid model.

[0122] To obtain the torque current I at the current moment q,t and rotor flux ψ r,t Then, the motor output torque T at the current moment e,t It can be calculated using the following formula:

[0123]

[0124] In the formula, G t (I,ω) and T e,t n represents the current motor output torque value. p I represents the number of pole pairs of the motor. q,t ψ represents the torque current at the current moment. r,t This represents the rotor flux linkage at the current moment.

[0125] Comparative Example 1

[0126] Based on Example 4, the mechanism-LSTM neural network hybrid model is compared and verified with the spindle cutting torque prediction method based on the LSTM neural network model disclosed in the invention patent with publication number CN117733646A.

[0127] In this embodiment, the experiment was conducted on a DMG DMU 80P machining center, and the experimental scenario was as follows. Figure 2As shown; the sampling frequency of the spindle servo monitoring signal and the spindle cutting torque is 2000Hz; two carbide end mills of the same specification with a diameter of 20mm and three cutting edges were used in the experiment; one was a new tool and the other was a tool that had suffered single-edge breakage during machining. Eight sets of linear milling experiments with varying cutting parameters were conducted. The tool status and cutting parameter settings are shown in Table 1, and a total of 148 trajectory data were obtained. The 148 trajectory data generated by the milling experiment were divided into three datasets and compared with different numbers. The datasets compare the predictive performance of the LSTM neural network model and the mechanistic-LSTM neural network hybrid model. Dataset 1 consists of all experimental data at a spindle speed of 1500 rpm, where groups 1 and 3 are the training set, and groups 5 and 7 are the test set. Dataset 2 consists of all experimental data at a spindle speed of 2000 rpm, where groups 2 and 4 are the training set, and groups 6 and 8 are the test set. Dataset 3 consists of all experimental data, where groups 1-4 are the training set, and groups 5-8 are the test set.

[0128] Table 1 Tool status and cutting parameter settings in the milling experiment

[0129]

[0130] The parameters of the motor torque estimation model are set as follows: number of motor pole pairs n p The value is 3, and the rotor resistance R is 3. r The Ω is 0.088, and the rotor inductance L is... r The mutual inductance between the stator and rotor is 0.0180H. m The value is 0.0193H, used to construct a motor torque estimation model based on a current-type flux observer; the hidden layer of the mechanism-LSTM neural network hybrid model is set to 512 dimensions, followed by a 512-dimensional to 1-dimensional linear fully connected layer for output; the initial state is set to a zero vector, and the training method is sequence-to-sequence; the LSTM neural network settings for comparison are the same as those of the mechanism-LSTM neural network hybrid model.

[0131] Table 2 Comparison of cutting torque prediction accuracy between the two models

[0132]

[0133]

[0134] Table 2 shows a comparison of the prediction accuracy of the LSTM neural network model and the mechanism-LSTM neural network hybrid model disclosed in this patent on different datasets. Test results under different datasets and tooling conditions are then visualized. The results are as follows: Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 As shown in Table 2; MAE refers to Mean Absolute Error, RRMSE refers to Root Mean Square Error, and CC refers to Correlation Coefficient. Figures 3 to 8 The labels in the table refer to the measurement data results of the cutting torque; from Table 2 and... Figures 3 to 8 As can be seen, the mechanism-LSTM hybrid neural network model shows significant improvement on different datasets; for example... Figure 7 and Figure 8 As shown, for the prediction of cutting torque under tool breakage state in dataset 3, which includes two rotational speeds, the LSTM neural network model performs poorly, while the mechanism-LSTM neural network model still performs well, and the difference between the two is very significant. Therefore, compared with the pure data-driven model represented by the LSTM neural network disclosed in the invention patent with publication number CN117733646A, the mechanism-LSTM hybrid neural network model disclosed in this embodiment can significantly improve the prediction effect of spindle cutting torque.

[0135] The above description is a detailed description of the preferred embodiments of this application. However, the embodiments are not intended to limit the scope of the patent application of this application. All equivalent changes or modifications made under the technical spirit of this application should fall within the patent scope covered by this application.

Claims

1. A method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque, characterized in that, Using the three-phase current of the spindle, the spindle speed, and the spindle angular acceleration as inputs, and the spindle cutting torque as output, a recurrent neural network model is constructed as the basic model. Then, the motor torque estimation model is used as the input of the basic model to construct a mechanism-data hybrid model. The recurrent neural network model is either a gated recurrent unit neural network model or a long short-term memory neural network model.

2. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 1, characterized in that: The motor output torque is obtained through a motor torque estimation model, and then used as the input to a mechanism-data hybrid model. The specific steps for solving the motor output torque are as follows: S1: The stator excitation I is obtained by transforming the three-phase current of the main shaft using a motor torque estimation model. d and stator torque current I q ; S2: Stator excitation I obtained from S1 d Find the rotor flux linkage ψ r And based on rotor flux ψ r and stator torque current I q The calculated speed difference ω sl Finally, the obtained speed difference ω sl The rotor flux linkage angle θ is obtained; S3: Based on the stator torque current I obtained in step S2 q and rotor flux ψ r The output torque of the motor is calculated and used as the input to the mechanism-data hybrid model.

3. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 2, characterized in that: The motor torque estimation model is used as the input to the gated recurrent unit neural network model, and the specific formula is expressed as follows: In the formula, T m Represents the spindle cutting torque, GRU represents the gated loop unit neural network model, and I represents the three-phase spindle current [I u ,I v ,I w ], ω represents the spindle speed, The model represents the spindle angular acceleration, h represents the hidden state of the model, t represents the current time, and t-1 represents the previous time; G represents the motor torque estimation model, which takes the three-phase current of the spindle and the spindle speed as inputs. t (I,ω) represents the motor output torque value at the current moment.

4. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 2, characterized in that: The motor torque estimation model is used as input to the long short-term memory neural network model, and the specific formula is expressed as follows: In the formula, T m The spindle cutting torque is represented by LSTM, the long short-term memory neural network model is represented by I, and the three-phase current of the spindle is represented by [I]. u ,I v ,I w ], ω represents the spindle speed, Let represent the spindle angular acceleration, h represent the hidden state of the model, c represent the element state of the model, t represent the current time, t-1 represent the previous time, and G represent the motor torque estimation model, which takes the three-phase current and speed of the spindle as inputs. t (I,ω) represents the motor output torque value at the current moment.

5. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 3 or 4, characterized in that: The motor torque estimation model adopts the motor torque estimation model of the current-type rotor flux observer.

6. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 5, characterized in that, The specific method of step S1 is as follows: the motor torque estimation model of the current-type rotor flux observer is used to estimate the three-phase current I. u I v and I w The stator excitation I is obtained by performing 3s / 2s and 2s / 2r transformations. d and stator torque current I q Its transformation formula is: In the formula, I d I represents the excitation of the stator. q C represents the stator torque current. 2s / 2r Represents the 2s / 2r transformation, C 3s / 2s This represents the 3s / 2s transformation, and θ represents the rotor flux linkage angle.

7. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 6, characterized in that: According to the stator excitation I d and stator torque current I q Find the rotor flux linkage ψ r The specific formula is: In the formula, τ r L is the rotor time constant of the motor. m Let be the stator inductance, and s represent the differential operator.

8. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 7, characterized in that: According to the rotor flux ψ r and stator torque current I q The speed difference ω was calculated. sl The specific formula is: In the formula, τ r L is the rotor time constant of the motor. m For stator inductance, τ r Let ψ be the rotor time constant of the motor, and ψ be the rotor flux linkage. r .

9. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 8, characterized in that: Through the speed difference ω sl The specific formula for obtaining the rotor flux linkage angle θ is: In the formula, ω r The expression represents the product of the actual speed of the motor and the number of pole pairs of the motor, and s represents the differential operator.

10. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 9, characterized in that: The product ω of the actual speed of the motor and the number of pole pairs of the motor. r The specific calculation formula is as follows: oh r =ωn p ; In the formula, ω represents the actual speed of the motor, and n p This indicates the number of pole pairs of the motor.

11. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 10, characterized in that: The final motor output torque obtained in step S3 is specifically formulated as follows: In the formula, T e n represents the output torque of the motor. p I represents the number of pole pairs of the motor. q Represents torque current, ψ r L represents the rotor flux linkage. r L represents the rotor inductance. m It is the stator inductance.

12. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 11, characterized in that: When it is necessary to obtain the stator torque current I at the current moment q,t and the rotor flux linkage ψ of the stator r,t The specific formula for calculating the motor output torque at the current moment is as follows: In the formula, G t (I,ω) and T e,t n represents the current motor output torque value. p I represents the number of pole pairs of the motor. q,t ψ represents the torque current at the current moment. r,t L represents the rotor flux linkage at the current moment. r L represents the rotor inductance. m It is the stator inductance.

13. The method for constructing a mechanism-data hybrid model for online prediction of machine tool spindle cutting torque according to claim 3 or 4, characterized in that: The inputs to the mechanism-data hybrid model in step S3 include the motor output torque, the three-phase current of the spindle, the spindle speed, and the spindle angular acceleration; the output of the mechanism-data hybrid model is the spindle cutting torque.

Images