阅读说明：本技术 数据驱动的数控车削元动作能耗预测方法 (Data-driven numerical control turning element action energy consumption prediction method ) 是由 李聪波 尹誉先 肖溱鸽 赵希坤 龙云 潘建 孙鑫 胡曾明 于 2019-10-14 设计创作，主要内容包括：本发明公开了一种数据驱动的数控车削元动作能耗预测方法,包括以下步骤：分析数控车削加工过程元动作组成,分析数控车削元动作能耗特性,揭示不同时段元动作对机床能耗影响；利用传感器及工业物联网技术对数控车削元动作进行判断及获取,基于元动作数据对机床状态进行判断；基于高斯过程回归算法,利用数据驱动方式建立数控车削元动作能耗预测模型,运用多种评价指标对所生成的模型进行评价。(The invention discloses a data-driven numerical control turning element action energy consumption prediction method, which comprises the following steps: analyzing the element action composition in the process of numerical control turning, analyzing the energy consumption characteristic of the element action of numerical control turning, and revealing the influence of the element action in different time periods on the energy consumption of the machine tool; judging and acquiring numerical control turning element actions by using a sensor and an industrial Internet of things technology, and judging the state of a machine tool based on element action data; based on a Gaussian process regression algorithm, a numerical control turning element action energy consumption prediction model is established in a data driving mode, and various evaluation indexes are used for evaluating the generated model.)

1. The data-driven numerical control turning element action energy consumption prediction method is characterized by comprising the following steps of:

step 1: analyzing the element action composition in the process of numerical control turning, analyzing the energy consumption characteristic of the element action of numerical control turning, and revealing the influence of the element action in different time periods on the energy consumption of the machine tool;

step 2: judging and acquiring numerical control turning element actions by using a sensor and an industrial Internet of things technology, and judging the state of a machine tool based on element action data;

and step 3: and establishing a numerical control turning element action energy consumption prediction model by using a data driving mode based on a Gaussian process regression algorithm.

2. The method for predicting the energy consumption of the data-driven numerical control turning element action according to claim 1, wherein the method comprises the following steps: in the step 1, a numerical control turning process element action set MA is represented as:

MA＝{MS,SR,AF,CF,CR,TC,C,MC,DC,PC}

in the formula, MS is a machine tool start-stop element action, SR is a main shaft rotating element action, AF is a feed shaft operating element action, CF is a cooling liquid start-stop element action, CR is a chip removal start-stop element action, TC is a cutter replacing element action, C is a cutting machining element action, MC is a workpiece material changing element action, DC is a machining diameter changing element action, and PC is a process parameter changing element action.

3. The method for predicting the energy consumption of the data-driven numerical control turning element action according to claim 1, wherein the method comprises the following steps: the judgment and acquisition process of the numerical control turning element action in the step 2 is as follows:

by communicating with the numerical control system, the position coordinate (X-axis coordinate X) of the machine tool cutter can be obtained_pZ axis coordinate Z_p) Data such as process parameters (main shaft rotating speed n and feeding rate f), a cooling liquid opening state, a chip removal opening state and the like; the total power P of the machine tool can be obtained by externally connecting a power sensor_totalAnd main transmission system power P_sp(ii) a Processing task information such as workpiece materials and the like can be acquired through the MES system; the search range of the data collected by the machine tool in real time can be greatly reduced by extracting the starting time and the ending time of the processing task in the MES system; by collecting and analyzing the data, the acquisition of the turning element action can be realized, and the method comprises the following steps:

(1) the machine tool opening element action MS comprises states of a first machine tool opening, a second machine tool opening and the like; when P is present_totalWhen the value is 0, the machine tool is in a stop state; when P is present_total>And 0, the machine tool is in an open state. Different machine tools have own numbers, and the machine tool opening element action state of each machine tool can be judged according to the numbers of the machine tools;

(2) the spindle rotating element acts SR, which comprises a static state and a rotating state, when n is equal to 0, the SR is in the static state; when n >0, SR is in a rotated state;

(3) the feed shaft operation element action AF comprises four states of static, X-axis feed, Z-axis feed and X-Z linkage feed, and when f is equal to 0, the AF is in a static state; when f is>0、X_pChange, Z_pWhen not changed, AF is in X-axis feeding state; when f is>0、X_pUnchanged, Z_pWhen the change occurs, the AF is in a Z-axis feeding state; when f is>0、X_pChange, Z_pWhen the change occurs, the AF is in an X-Z linkage feeding state;

(4) the cutting fluid start and stop element acts CF, including closing and opening two states, and the CF state can be directly obtained by communicating with a numerical control system;

(5) the chip removal start-stop element acts CR, including closing and opening two states, and the CR state can be directly obtained by communicating with the numerical control system;

(6) the tool replacing element action TC comprises a state of replacing a tool I, a state of replacing a tool II and the like, wherein each tool has a corresponding tool number on a tool rest, the tool number currently used by the machine tool can be obtained by communicating with a numerical control system, and the TC state can be judged when the read tool number changes;

(7) the cutting element acts C, including non-cutting and cutting, when P is_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u>C₀When the cutting tool is used, C is in a cutting state; otherwise, C is in an uncut state. Wherein P is_uFor no-load power of the machine tool, C₀According to the cutting amount, the P after the main shaft is started is generally 5% -10%_spReal-time storage of array [ P ]_sp]In, get stable [ P ]_sp]Average value of_u；

(8) The workpiece material change element action MC comprises a first material state, a second material state and the like, and the information of the processed workpiece material can be directly obtained through an MES system;

(9) the machining diameter change element action DC comprises a size change state of the machining diameter D and an X-axis coordinate X of the position of the cutter_pThe value of (D) is the size of the processing diameter D;

(10) worker's toolProcess parameter variation element action PC comprising cutting speed v_cVariation, feed rate f variation and back draft a_pVariation, cutting speed v_cFrom the formula v_cCalculating out pi nD; the feed rate f can be directly obtained by communicating with a numerical control system; back draft a_pEqual to the difference between the radius of the part and the machining radius of the first cutting machining or the difference between the machining radii of the two cutting machining.

4. The method for predicting the energy consumption of the data-driven numerical control turning element action according to claim 1, wherein the method comprises the following steps: the state judgment method based on the numerical control turning element action data in the step 2 comprises the following steps:

(1) standby state determination

The machine tool only maintains the most basic operation in the state, the judgment of the standby state can be realized by monitoring the power of the machine tool when P is_total>When the state of the MS is in an opening state, and when the states of the SR and AF are in a static state, the running state of the machine tool is in a standby state;

(2) determination of idle cut state

In this state the feed system of the machine tool is started and the auxiliary system is switched on and the machine tool does not start cutting the workpiece, when f>At 0, AF is in a certain axial feeding state, and when P is_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u<C₀When the C is in an uncut state, the running state of the machine tool is in a free cutting state;

(3) cutting state judgment

In this state, the machine tool cuts off the redundant material of the workpiece, so that the cutting element action of the machine tool is added besides the element action of the idle cutting time interval in the state, when P is_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u>C₀When the C is in the cutting state, namely, the machine tool is judged to be in the cutting state;

(4) tool change state element action

And in the state, the machine tool replaces the cutter for processing the part, and the machine tool is in a cutter replacing state when the TC generates a cutter replacing action by reading the number of the currently used cutter of the machine tool.

5. The method for predicting the energy consumption of the data-driven numerical control turning element action according to claim 1, wherein the method comprises the following steps: in step 3, a Gaussian process regression algorithm is utilized to establish a data-driven numerical control turning element action energy consumption prediction model, which can be expressed as:

where E is an observed value of energy consumption affected by noise, and y_iFor machining the ith step power observation, t, for machine tools affected by noise_iFor step i, run time, x_iIs the meta-action data of step i, f (x)_i) For the theoretical value of the power of the ith step and obeying the gaussian distribution,

Technical Field

The invention relates to the field of mechanical cutting machining, in particular to a numerical control turning machining energy consumption prediction method.

Background

Machine tools, which are the main manufacturing equipment in the manufacturing industry, are energy-consuming but inefficient. And the machine tool is used as a complex machining manufacturing system, the energy characteristic composition is complex, and the accurate prediction of the energy consumption is difficult. The reliable machine tool energy consumption model can provide accurate energy consumption predicted values for processing under different processing conditions or different processing parameters, is a key for solving the problem of machine tool energy efficiency, and provides a basis for energy efficiency optimization. Existing research on energy consumption prediction considers a few factors, and is often conducted on specific machining operations, parameter spaces and tool-workpiece material combinations, which limits the generalization of energy consumption prediction methods. Therefore, the method starts from the basic action of the bottom layer of the machine tool, analyzes the energy consumption mechanism of each action of the machine tool, utilizes more and more machining process data of a workshop, establishes a numerical control turning energy consumption model based on Gaussian process regression by using a data-driven mode, and improves the accuracy and the generalization of energy consumption prediction. By the method, the energy consumption consumed by each action of the machine tool when the machine tool processes the parts can be obtained more carefully, more accurate part processing energy consumption can be obtained, and a more accurate energy consumption model is provided for researching machine tool parameter optimization, process route optimization, workshop scheduling and the like.

Disclosure of Invention

The invention provides a data-driven numerical control turning element action energy consumption prediction method, so as to obtain a more accurate energy consumption prediction model.

The technical scheme adopted for achieving the aim of the invention is that the method for predicting the energy consumption of the action of the numerical control turning element driven by data is characterized by comprising the following steps of:

step 1: analyzing the element action composition in the process of numerical control turning, analyzing the energy consumption characteristic of the element action of numerical control turning, and revealing the influence of the element action in different time periods on the energy consumption of the machine tool;

step 2: judging and acquiring numerical control turning element actions by using a sensor and an industrial Internet of things technology, and judging the state of a machine tool based on element action data;

and step 3: and establishing a numerical control turning element action energy consumption prediction model by using a data driving mode based on a Gaussian process regression algorithm.

Preferably, the element actions in the step 1 of the numerical control turning process can be divided into machine element actions and generalized element actions according to whether the machine part moves or not:

(1) the machine tool element actions comprise machine tool starting and stopping element actions, main shaft rotating element actions, feed shaft operating element actions, cooling liquid starting and stopping element actions, chip removal starting and stopping element actions, cutter replacing element actions and cutting machining element actions, and the part of element actions are all that machine tool parts can move and change in the machining process of a machine tool.

(2) The generalized element actions comprise a workpiece material change element action, a machining diameter change element action and a process parameter change element action, and although the part of element actions do not cause the machine tool to generate direct motion change, the part of element actions can affect the energy consumption of the whole machining process.

The numerical control turning process element action set MA can be expressed as:

MA＝{MS,SR,AF,CF,CR,TC,C,MC,DC,PC}

in the formula, MS is a machine tool start-stop element action, SR is a main shaft rotating element action, AF is a feed shaft operating element action, CF is a cooling liquid start-stop element action, CR is a chip removal start-stop element action, TC is a cutter replacing element action, C is a cutting machining element action, MC is a workpiece material changing element action, DC is a machining diameter changing element action, and PC is a process parameter changing element action. The specific classification of the numerical control turning element actions and the owned states of the element actions are shown in fig. 1.

Preferably, in step 1, the impact analysis of the energy consumption characteristics of the numerical control turning element actions and the actions in different time periods on the energy consumption is as follows:

various machining actions are divided into four time periods for analysis, so that the energy consumption generated by machining one part comprises the following components:

E＝E_st+E_air+E_c+E_tc

wherein E is_stFor standby period energy consumption, E_airFor idle cutting period energy consumption, E_cFor energy consumption during cutting, E_tcEnergy consumption is realized in the tool changing period. Each step of machining action when the part is machined by the lathe consists of element action, energy consumption of each machining action also consists of energy consumption generated by the element action, and element action composition and energy consumption analysis in different time periods are favorable for building a better modelAnd adding an accurate and reasonable energy consumption model. The cutting action and power curve of the numerical control turning is shown in fig. 2 and 3. The action composition and energy consumption analysis of each time interval element are as follows:

(1) standby time interval meta-action composition and energy consumption analysis

Energy consumption E in Standby period_stThe energy consumption of the lathe for maintaining normal operation after the lathe is started is as follows:

in the formula, E_bThe energy consumption is the basic energy consumption of the lathe, represented by the energy consumption generated after MS is started, and mainly comprises the energy consumed by basic components such as an illumination system, a display, a numerical control system, a servo system, a frequency converter and the like of the lathe; p_bThe basic power of the lathe; t is t_stIs the standby duration. CHEN finds that the energy consumption of basic parts of the machine tool accounts for about 21% by counting the energy consumption of the machine tool for ten days, so E_bIs a non-negligible component during processing. So the standby period energy consumption can also be expressed as:

E_st＝f(MS)

(2) idle-cut time interval element action composition and energy consumption analysis

Energy consumption E during idle cutting period_airThe energy consumption formula is that the energy consumed by each part of the lathe in the idle feed time period when the lathe is about to start cutting is as follows:

in the formula, E_spWork against friction during rotation of the spindle, the partial power P_spRelated to spindle motor losses, spindle speed, i.e. MS, SR, PC. E_fThe power P of the part is the work which is required to overcome the friction force in the X direction and the Y direction when the lathe is fed_fExpressed as a function related to feed motor losses, feed speed, feed direction, i.e. the fraction of power is related to MS, AF, PC. E_cf、E_crFor cooling pump motor and removing chipsEnergy consumed by the motor, power P of the portion_cf、P_crIs relatively constant and is related to the model of each motor configured by the lathe, namely MS, CF and CR. t is t_airThe length of the blank cut.

In summary, the energy consumption of this period is related to MS, SR, AF, CF, CR, PC, so the energy consumption of the idle-cut period can also be expressed as:

E_air＝f(MS,SR,AF,CF,CR,PC)

(3) cutting time interval element action composition and energy consumption analysis

Energy consumption in cutting period E_cThe energy consumption is represented by the energy consumed in the process of removing the workpiece material by the lathe, and the energy consumption composition formula is as follows:

in the formula, E_mTo remove energy consumption for the material, E_adEnergy is consumed for additional load. E_mRemoving power P of material for removing energy consumption of workpiece material by machine tool_mIs denoted as P_m＝F_cv_c＝k_cAnd (4) MRR. Wherein F_cAs a cutting force, v_cAs cutting speed, k_cFor the coefficient of cutting force, MRR is the material removal rate (for turning, the value is equal to the cutting speed v_cFeed rate f and back draft a_pThe product of (d). MRR is related to the cutting parameter, k_cRelating to workpiece material, tool nose arc radius, tool angle, cooling conditions, cutting direction, etc. E_adThe power P of the part is the load loss caused by the increase of cutting force and torque when the machine tool cuts_adAnd P_mIs approximately quadratic, so it can be considered that P is influenced_mAll will be on P_adAn influence is produced. t is t_cIs the cutting time. Therefore E_mAnd E_adThe influencing factors of (1) comprise MS, CF, TC, C, MC, DC and PC correlation.

In summary, E_cAssociated with MS, SR, AF, CF, CR, TC, C, MC, DC, PC, the cutting interval energy consumption can also be expressed as:

E_c＝f(MS,SR,AF,CF,CR,TC,C,MC,DC,PC)

(4) tool changing time interval element action composition and energy consumption analysis

Tool change period energy consumption E_tcThe energy consumed by automatic tool changing realized by rotating the tool rest after the represented lathe tool returns to the tool changing point is as follows:

in the formula, E_teAnd P_teThe energy and power consumed for tool change is related to the number of tool bits the tool changing motor rotates, i.e. to MS, TC. t is t_tcIs the tool change time. The formula is composed of energy consumption to obtain E_tcRelated to MS, SR, CF, CR, TC, PC. The tool change period energy consumption can also be expressed as:

E_c＝f(MS,SR,CF,CR,TC,PC)

preferably, in step 2, the method for judging and acquiring the numerical control turning element action by using the sensor and the industrial internet of things technology comprises the following steps:

by communicating with the numerical control system, the position coordinate (X-axis coordinate X) of the machine tool cutter can be obtained_pZ axis coordinate Z_p) Taking a FANUC system as an example, reading system internal data by using functions in a FOCAS function library, for example, reading the real-time rotation speed of a machine tool spindle by using a cnc _ acts () function, and reading the real-time absolute coordinates of the machine tool by using a cnc _ absolute () function; the total power P of the machine tool can be obtained by externally connecting a power sensor_totalAnd main transmission system power P_sp(ii) a The MES system can obtain the information of the processing task, such as the workpiece material. And the search range of the data collected by the machine tool in real time can be greatly reduced by extracting the starting time and the ending time of the processing task in the MES system, and the efficiency of data integration is improved. The meta-action acquisition flow is shown in fig. 5. Through collecting and analyzing the data, the acquisition of the turning element action can be realized, and the steps are as follows:

(1) machine tool opening element action MSThe method comprises the states of a machine tool with one opening, a machine tool with two openings and the like; when P is present_totalWhen the value is 0, the machine tool is in a stop state; when P is present_total>And 0, the machine tool is in an open state. Different machine tools have own numbers, and the machine tool opening element action state of each machine tool can be judged according to the numbers of the machine tools; (2) the spindle rotating element acts SR, which comprises a static state and a rotating state, when n is equal to 0, the SR is in the static state; when n is>At 0, SR is in a rotating state;

(3) the feed shaft operation element action AF comprises four states of static, X-axis feed, Z-axis feed and X-Z linkage feed, and when f is equal to 0, the AF is in a static state; when f is>0、X_pChange, Z_pWhen not changed, AF is in X-axis feeding state; when f is>0、X_pUnchanged, Z_pWhen the change occurs, the AF is in a Z-axis feeding state; when f is>0、X_pChange, Z_pWhen the change occurs, the AF is in an X-Z linkage feeding state;

(4) the cutting fluid start and stop element acts CF, including closing and opening two states, and the CF state can be directly obtained by communicating with a numerical control system;

(5) the chip removal start-stop element acts CR, including closing and opening two states, and the CR state can be directly obtained by communicating with the numerical control system;

(6) the tool replacing element action TC comprises a state of replacing a tool I, a state of replacing a tool II and the like, wherein each tool has a corresponding tool number on a tool rest, the tool number currently used by the machine tool can be obtained by communicating with a numerical control system, and the TC state can be judged when the read tool number changes;

(7) the cutting element acts C, including non-cutting and cutting, when P is_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u>C₀When the cutting tool is used, C is in a cutting state; otherwise, C is in an uncut state. Wherein P is_uFor no-load power of the machine tool, C₀According to the cutting amount, the P after the main shaft is started is generally 5% -10%_spReal-time storage of array [ P ]_sp]In, get stable [ P ]_sp]Average value of_u；

(8) The workpiece material change element action MC comprises a first material state, a second material state and the like, and the information of the processed workpiece material can be directly obtained through an MES system;

(9) the machining diameter change element action DC comprises a size change state of the machining diameter D and an X-axis coordinate X of the position of the cutter_pThe value of (D) is the size of the processing diameter D;

(10) process parameter variation element action PC, including cutting speed v_cVariation, feed rate f variation and back draft a_pVariation, cutting speed v_cFrom the formula v_cCalculating out pi nD; the feed rate f can be directly obtained by communicating with a numerical control system; back draft a_pEqual to the difference between the radius of the part and the machining radius of the first cutting machining or the difference between the machining radii of the two cutting machining.

Preferably, in step 2, the method for determining the input meta-motion data based on the machine state judgment of the meta-motion data and establishing the model includes:

in the machining process, the energy characteristics of the machine tool are not constant, and different motion states are completed by matching different element actions. By classifying the machine tool states and inputting different meta-motion data in different states, the data training number required by the model can be reduced, and the accuracy of the model is improved. Based on the state of the meta-motion in the cutting process, the operation state of the machine tool can be identified simply and quickly, and the method and the flow are shown in fig. 6.

(1) Standby state determination

In this state, the machine tool only maintains the most basic operation. The judgment of the standby state can be realized by monitoring the power of the machine tool when P is_total>When 0, MS state is in on state, and when SR, AF state is in static state, the running state of the machine tool is in standby state. According to the meta-action energy consumption analysis in the step 1, the variable needing to be input in the state is MS.

(2) Determination of idle cut state

In this state the feed system of the machine tool is started and the auxiliary system is switched on and the machine tool does not start cutting the workpiece. When f is>At 0, AF is in a certain axial feeding state, and when P is_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u<C₀And C is in an uncut state, and the running state of the machine tool is in a free-cutting state. According to the meta-action energy consumption analysis in step 1, the variables to be input in the state are { MS, SR, AF, CF, CR, DC, PC }.

(3) Cutting state judgment

In this state, the machine tool cuts off the redundant materials of the workpiece, so the machine tool cutting element action is added besides the element action of the idle cutting time interval. When P is present_total≠0，f≠0，n≠0，(P_sp-P_u)/P_u>C₀And C is in a cutting state, namely, the machine tool is judged to be in the cutting state. According to the meta-action energy consumption analysis in the step 1, the variable needing to be input in the state is { MS, SR, AF, CF, CR, TC, C, MC, DC, PC }.

(4) Tool change state element action

In this state, the machine tool replaces the tool for machining the part. By reading the number of the currently used tool of the machine tool, when the TC generates a tool replacing action, the machine tool is in a tool changing state. According to the meta-action energy consumption analysis in the step 1, the variable needing to be input in the state is { MS, SR, CF, CR, TC, PC }.

Preferably, in step 3, based on a gaussian process regression algorithm, the process of establishing the numerical control turning element action energy consumption prediction model by using a data driving mode is as follows:

the Gaussian Process Regression (GPR) model is a machine learning algorithm developed based on Bayesian theory and statistical theory, has good adaptability to processing complex problems of high dimensionality, small samples, nonlinearity and the like, has strong generalization capability, and has the advantages of easy implementation, super-parameter adaptive acquisition, flexible nonparametric inference, probability significance of output and the like compared with a neural network and a support vector machine. In view of this, the present document uses a gaussian process regression algorithm to input corresponding machine tool machining process meta-motion states for training in different machine tool states. The machine tool energy consumption gaussian process regression model can be expressed as:

where E is an observed value of energy consumption affected by noise, and y_iFor machining the ith step power observation, t, for machine tools affected by noise_iFor step i, run time, x_iIs the meta-action data of step i, f (x)_i) For the theoretical value of the power of the ith step and obeying the gaussian distribution,

and m is the random variable noise of the ith step, and the motion step number of the machine tool machining.

The evaluation process of the generated model by using various evaluation indexes comprises the following steps:

to understand how each individual energy consumption prediction model performs under invisible test data, the error of each prediction model is estimated using k-fold cross validation (k-fold cross validation). Firstly, a sample is randomly divided into k disjoint parts, wherein one part serves as a training data set, and the rest serves as a test set for verifying the performance of the model. And then repeating the moving training set and the test set k times, calculating the performance index in each cycle, and taking the average value of the k times of evaluation to evaluate the prediction effect of the model. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), mean relative error (MAPE), and coefficient of solution (R) are used herein²) As an evaluation index, the calculation formula is as follows:

where M is the number of samples, y_iAs an observed value of the ith energy consumption of the sample,is a predicted value of the ith energy consumption. RMSE, MAE, MAPE can be used to quantify the absolute error of a predictor from an observed value, with smaller values indicating closer proximity of the predictor to the observed value. R²The reliability of the model is evaluated, and the value range is [0,1 ]]In between, a value closer to 1 indicates higher model prediction accuracy.

Drawings

FIG. 1 is a schematic diagram of a numerical control turning process

FIG. 2 numerical control turning process

FIG. 3 numerical control turning power curve

FIG. 4 data-driven numerical control turning element action energy consumption modeling flow

FIG. 5 Meta action acquisition flow

FIG. 6 State judgment and input variable determination based on Meta-actions

FIG. 7 data acquisition hardware and software platform

FIG. 8 is a comparison graph of observed values and predicted values of training set and test set

FIG. 9 normalized residual histogram and scatter plot

FIG. 10 comparison of before and after processing of the workpiece

FIG. 11 comparison graph of observed value and predicted value of energy consumption in the processing procedure

Detailed Description

The present invention will be further described with reference to the accompanying drawings and examples, but it should not be construed that the scope of the above-described subject matter is limited to the examples. Various substitutions and alterations can be made without departing from the technical idea of the invention and the scope of the invention is covered by the present invention according to the common technical knowledge and the conventional means in the field.

In the case, a machining workshop of a limited liability public driver of Chongqing God agricultural equipment is taken as a platform, a software and hardware integrated machine tool energy efficiency monitoring system independently developed by Chongqing university is used for acquiring real-time power data of a machine tool and internal data of a machine tool numerical control system, and a machining task information is acquired by a machining manufacturing workshop energy efficiency optimization and promotion support system independently developed by Chongqing university.

The real-time on-line monitoring of the power is realized through an HC33C3 type power sensor, the equipment respectively collects the total power of the machine tool and the power of a main transmission system (the collection frequency is 4HZ) at a main power supply of the machine tool and a main shaft servo system, and an intelligent terminal collects data of the power sensor through a Modbus communication protocol to realize real-time monitoring of the total power of the machine tool and the power of the main shaft. The internal data of the numerical control system is obtained by communicating with the numerical control system, taking a FANUC system as an example, the method uses a PCMCIA network card to realize hardware connection with an NC system, and the internal data of the system is read by using functions in a FOCAS function library. The intelligent terminal transmits the production process data to the server in real time through a TCP/IP protocol, and workers can check the production process data in real time through the MES system and also can transmit the processing task information to the intelligent terminal of each machine tool through the MES system. The hardware and software platform for data acquisition is shown in fig. 7. Three numerically controlled lathes, three common cutters and three common workpiece materials (45, Al and HT250) in a workshop are selected for research, and the method specifically comprises the following steps:

TABLE 1 machine tool types and parameters

TABLE 2 tool types and parameters

The method comprises the following steps:

the ten analyzed actions exist in the three numerical control machines of the case in the machining process, so the element action set MA in the numerical control turning process can be expressed as follows:

MA＝{MS,SR,AF,CF,CR,TC,C,MC,DC,PC}

and selecting the processing task information of the workshop parts through an MES system, and collecting more than 1000 pieces of data for research by depending on the software and hardware. Firstly, machine tool state judgment is carried out on each piece of data, and corresponding variable input models are selected to be trained under different states, and the specific method is as follows:

(1) when the MS state is in an opening state, and the SR state and the AF state are in a static state, the running state of the machine tool is in a standby state, and the variable needing to be input in the state is MS;

(2) when AF is in a feeding state of a certain axis and C is in a non-cutting state, the running state of the machine tool is in a non-cutting state, and variables needing to be input in the state are { MS, SR, AF, CF, CR, DC, PC };

(3) when the C is in a cutting state, namely the machine tool is judged to be in the cutting state, and variables needing to be input in the state are { MS, SR, AF, CF, CR, TC, C, MC, DC and PC };

(4) when the TC generates a cutter replacing action, the machine tool is in a cutter changing state, and variables needing to be input in the state are { MS, SR, CF, CR, TC and PC }.

Method for dividing training set for sample data by using cross validation method during model training

And test set

And evaluated, in this studyAnd

the data volume ratio was 4: 1.

The data are input into a Gaussian process regression model for training, and mu (x | D) is calculated_q) And σ²(x|D_q) The model f (x) is obtained. Response Surface Methods (RSM) and Artificial Neural Networks (ANN) are two of the most common modeling methods, and in order to verify the effect of the GPR algorithm, the two modeling methods are selected for comparative analysis with GPR. RSM loopThe model is determined by adopting a multivariate quadratic regression equation, and equation coefficients are solved by a least square method, so that a functional relation between factors and response values is fitted; the ANN is composed of an input layer, a hidden layer and an output layer, each layer comprises a large number of nodes, each node has a corresponding output function, the nodes of each layer are mutually connected to transmit information, and the input-output relation can be obtained by selecting the proper number of hidden layers, the number of nodes, the output function and the weight of each node. The evaluation index results of the training set and the prediction set of each method are shown in table 3.

TABLE 3 evaluation index results for training set and prediction set

As can be seen from Table 3, each index of GPR is superior to RSM and ANN, which indicates that GPR has the best model prediction effect, wherein each index of RSM is the worst, indicating that the model prediction effect is the worst. And all GPRs have less than 5% MAPE, R²The numerical control turning energy consumption can be effectively predicted by the model when the numerical control turning energy consumption is larger than 95%. The evaluation indexes of all algorithm training sets are superior to those of the test set. A group of training sets and testing sets are randomly taken to draw a comparison graph of the observed values and the predicted values, in order to overcome unreasonable influence on analysis results caused by different units, the energy consumption predicted values are normalized and are arranged from small to large according to the size of the predicted values, and the observed values and the predicted values of the training sets and the testing sets are compared as shown in the figure 8.

The comparison graph of the observed value and the predicted value shows that the three algorithms can better predict the energy consumption of the machine tool, the red line in the graph represents the predicted energy consumption value of each regression algorithm, and the blue point is the observed value of the energy consumption. Other algorithms predict by fitting a function, and GPR predicts by obtaining a probability distribution for each value, so GPR can calculate an empirical confidence interval, where the gray region in the graph represents the energy consumption prediction confidence interval with 95% confidence (μ (x | D)_q)±1.96σ²(x|D_q)). The graph can intuitively find that most blue points are in the range of the gray area, which shows that most observed values are in the 95% confidence range of energy consumption prediction, and the effectiveness of the model is reflected; the wider phenomenon appears in the partial confidence interval, which indicates that the data volume is less and the knowledge and experience of the model are insufficient.

In order to further study the prediction effects of the three regression algorithms, the test set data in the energy consumption data sample is taken for residual analysis, and the normalized residual histogram and the scatter plot are shown in fig. 9. The normalized residual histogram shows that the residuals of the models established by the three algorithms all have the trend of obeying normal distribution, which shows that the three models are reasonable and appropriate; the normalized residual histogram shows that most of the normalized residuals of the models established by the three algorithms are within a (-2, 2) interval, which indicates that the residuals meet the requirement of normal distribution. In the normalized residual histogram and the scatter diagram, the residuals of the GPR model are more concentrated to 0, which also indicates that the fitting accuracy of the model established by the GPR algorithm is higher than that of the models established by the RSM and the ANN.

And establishing a theoretical energy consumption model during a turning process by using the first machine tool, the first cutter and the first material by using a theoretical modeling method. The collected data including the process conditions were used to perform model evaluation, and the evaluation values are shown in table 4. As can be seen from the table, each evaluation index of the theoretical model is inferior to that of the Gaussian model. As the theoretical modeling is used for fitting the correlation coefficient through experiments, the experimental data are less, and the accuracy of the established model is lower. And because the machining process is a dynamic change process, the traditional machine tool energy consumption model and the motion of the machine tool are relatively determined, and after the process conditions are changed, a large number of physical parameters need to be modified to enable the model to adapt to new conditions. And the energy consumption model is driven by the data based on the element action, so that the action state and the change of the machine tool can be tracked in real time, the dynamic prediction of the machine tool energy consumption model is realized, and the energy consumption of any workpiece or any procedure can be predicted.

TABLE 4 comparison of evaluation indexes of Gaussian model and theoretical model

Fig. 10 is a comparison of before and after processing of the workpiece, and the processing conditions of the workpiece are shown in table 5. The research takes turning of the following two workpieces as an example, the energy consumption value of each step of action of a machine tool in the processing process of the two workpieces is obtained in real time through an intelligent terminal, and the energy consumption of each step of action in the processing process of the two workpieces is predicted through an established numerical control turning energy consumption dynamic prediction model. Because the idle cutting and the rapid feeding are short, the generated energy consumption is smaller than that of the cutting process, the part of energy consumption is merged when a comparison graph is drawn, and the comparison graph of the observed value and the predicted value of the energy consumption in the workpiece machining process is shown in fig. 11.

TABLE 5 workpiece processing conditions

As can be seen from the figure, the numerical control turning energy consumption prediction model established in the foregoing can substantially and accurately predict the energy consumption of each step of the machine tool. The difference between the predicted value and the observed value of the energy consumption in the second step of fig. 11 is large, and analysis shows that the oxide skin on the surface of the workpiece needs to be removed in the cutting process in the step, so that the cutting power is increased during the oxide skin cutting, and the energy consumption in the step is increased; the reason why the energy consumption prediction effect in the fast feed and idle feed states is not good in the cutting state is that the fast feed and idle feed time is short and the data fluctuation is large.