WO2020220221A1 - 生产设备的加工参数设置方法、装置和计算机可读介质 - Google Patents

生产设备的加工参数设置方法、装置和计算机可读介质 Download PDF

Info

Publication number
WO2020220221A1
WO2020220221A1 PCT/CN2019/085055 CN2019085055W WO2020220221A1 WO 2020220221 A1 WO2020220221 A1 WO 2020220221A1 CN 2019085055 W CN2019085055 W CN 2019085055W WO 2020220221 A1 WO2020220221 A1 WO 2020220221A1
Authority
WO
WIPO (PCT)
Prior art keywords
quality inspection
workpiece
production equipment
workpieces
quality
Prior art date
Application number
PCT/CN2019/085055
Other languages
English (en)
French (fr)
Inventor
冯程
施内加斯丹尼尔
曲颖
田鹏伟
Original Assignee
西门子股份公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西门子股份公司 filed Critical 西门子股份公司
Priority to US17/606,774 priority Critical patent/US11982996B2/en
Priority to CN201980091385.8A priority patent/CN113646714A/zh
Priority to PCT/CN2019/085055 priority patent/WO2020220221A1/zh
Priority to EP19927216.2A priority patent/EP3951523A4/en
Priority to EP23200043.0A priority patent/EP4276554A3/en
Publication of WO2020220221A1 publication Critical patent/WO2020220221A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/4183—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by data acquisition, e.g. workpiece identification
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
    • G05B19/401—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by control arrangements for measuring, e.g. calibration and initialisation, measuring workpiece for machining purposes
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
    • G05B19/408—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by data handling or data format, e.g. reading, buffering or conversion of data
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
    • G05B19/4155—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by programme execution, i.e. part programme or machine function execution, e.g. selection of a programme
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32187—Correlation between controlling parameters for influence on quality parameters
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32193—Ann, neural base quality management
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/34—Director, elements to supervisory
    • G05B2219/34295—System, logic analyser, simulation
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/42—Servomotor, servo controller kind till VSS
    • G05B2219/42152—Learn, self, auto tuning, calibrating, environment adaptation, repetition

Definitions

  • the present invention relates to industrial big data technology, in particular to a workpiece data processing method, device and computer readable medium.
  • the quality of the workpiece processed by a production equipment largely depends on the parameter settings of the production equipment.
  • CNC machine tools CNC machine tools
  • the setting of CNC machining parameters such as: cutting position, cutting angle, shaft height, and clamping position, will cause changes in the quality attribute value of the processed workpiece.
  • a workpiece that has passed quality must pass quality inspection, that is, its quality attribute value needs to pass in the quality inspection (usually, different types of workpieces need to be inspected for different quality attributes, for example, for worm gears in the automotive industry. Check the average torque, transmission ratio, peak speed, etc.). Therefore, the setting of processing parameters of production equipment is very important for workpiece manufacturing, especially for the optimization of workpiece quality in the finishing industry.
  • the processing parameters of production equipment such as CNC are usually set based on human experience, and the optimal setting of parameters may be affected by the acceptable range of various quality attributes of the workpiece. Therefore, it is often difficult to achieve only relying on human experience.
  • Optimal settings of processing parameters In addition, the quality of the workpiece is easily affected by the surrounding environmental conditions, such as the temperature and humidity during the manufacturing process of the workpiece. Therefore, when setting the processing parameters, the changes in the surrounding environmental conditions must be considered, which increases the difficulty of setting the processing parameters.
  • the quality of the workpiece is related to the setting of the processing parameters of the production equipment, and it is difficult to optimize the quality of the workpiece by relying only on human experience to set the processing parameters.
  • the quality of workpieces is often measured by quality inspection, so how to set the processing parameters of production equipment to improve the pass rate of workpiece quality inspection is an urgent problem to be solved.
  • One possible method is to realize the automatic optimal setting of the processing parameters of the production equipment by determining the relationship between the processing parameters of the production equipment and the pass rate of the workpiece quality inspection.
  • Machine learning algorithms can be used to determine the relationship between the processing parameters of the workpiece production equipment and the pass rate of the workpiece quality inspection.
  • One way is to use a supervised learning algorithm to build a model and determine the relationship by learning a large amount of labeled data.
  • data collection and labeling are often difficult and costly. This situation is particularly prominent in undata-based factories.
  • To determine the relationship between the processing parameters of a production equipment in the manufacturing process and the quality inspection pass rate of its processed workpieces often requires a lot of data.
  • the mass pass rate of the worm wheel under the following CNC machining parameter settings: the y-axis cutting position is 91, the helix angle is -2.97, the z-axis height is 209.8, the ambient temperature is 7 degrees Celsius, and the humidity is 70%.
  • the completion of the above processing usually requires the collection of hundreds of worm gear samples to ensure statistical significance, so that the accurate worm gear quality inspection pass rate can be calculated under a single combination of CNC machining parameters and surrounding environmental conditions.
  • For each combination of CNC processing parameters and surrounding environmental conditions there are often only a few labeled data. It is difficult to obtain the data under different surrounding environmental conditions with the help of traditional regression models. Accurate relationship between CNC machining parameters and workpiece quality inspection results.
  • the embodiments of the present invention provide a workpiece data processing method, device, and computer readable medium, which are used to accurately determine the relationship between the processing parameters of the production equipment, the data of the surrounding environmental conditions, and the quality inspection result of the workpiece.
  • the optimal processing parameters can be determined based on the relationship.
  • a method for processing workpiece data including: acquiring processing condition data, quality attribute values, and quality inspection result data for each of a plurality of workpieces processed by a production equipment, wherein the processing condition data of one workpiece includes : The processing parameters used by the production equipment when processing the workpiece and the surrounding environmental condition data when the production equipment is processing the workpiece; determine according to the processing condition data and quality attribute values of each of the plurality of workpieces The first relationship between the quality attribute value of the workpiece processed by the production equipment and the ambient environmental condition data when the production equipment processes the workpiece and the processing parameters of the production equipment; according to the value of each of the plurality of workpieces The quality attribute value and the quality inspection result data determine the second relationship between the quality inspection result data and the quality attribute value of the workpiece processed by the production equipment.
  • a workpiece data processing device including: a workpiece data acquisition module configured to acquire processing condition data, quality attribute values, and quality inspection result data of each of a plurality of workpieces processed by a production equipment ,
  • the processing condition data of one of the workpieces includes: the processing parameters used by the production equipment when processing the workpiece and the surrounding environment condition data when the production equipment is processing the workpiece;
  • a first relationship determination module is configured to According to the processing condition data and quality attribute value of each of the plurality of workpieces, the quality attribute value of the workpiece processed by the production equipment and the surrounding environmental condition data when the production equipment processes the workpiece and the quality attribute value of the production equipment are determined A first relationship between processing parameters;
  • a second relationship determination module configured to determine the quality inspection result of the workpiece processed by the production equipment according to the quality attribute value and quality inspection result data of each of the plurality of workpieces The second relationship between the data and the quality attribute value.
  • the quality attribute values of the workpiece processed by the production equipment and the processing parameters of the production equipment are determined by processing the workpiece data , The first relationship between the surrounding environmental conditions when processing the workpiece, and the second relationship between the workpiece quality attribute value and the workpiece quality inspection result data.
  • each of the plurality of workpieces processed by the production equipment may be acquired The processing condition data, the quality attribute value and the quality inspection result data, so that all the acquired data covers as much as possible the combination of various surrounding environmental condition data and processing parameters. In this way, for each combination of the processing parameters of the production equipment and the surrounding environmental conditions, only a small amount of data can be used to train the mathematical model of the first relationship.
  • a Conditional Generative Adversarial Net (CGAN) model may be used to obtain the first relationship between the distribution of the processing parameters of the production equipment and the quality attributes of the workpiece under different ambient conditions.
  • the processing condition data of each of the plurality of workpieces can be used as an input vector of a generator in a CGAN model; the processing condition data of each of the plurality of workpieces can be used as the judgment of the CGAN model
  • An input vector of the CGAN model use the quality attribute value of each of the plurality of artifacts or the output vector of the generator of the CGAN model as another input vector of the discriminator in the CGAN model; train the The CGAN model uses the generator of the CGAN model obtained by training as the first relationship.
  • the trained CGAN model can generate any number of simulated workpiece samples. Using the method provided by the embodiment of the present invention can significantly reduce the data required to determine the relationship between the processing parameters and the result of the workpiece quality inspection.
  • the quality attribute value of each of the plurality of first simulated workpieces can be generated according to the processing condition data, the quality attribute value of each of the plurality of workpieces processed by the production equipment, and the first relationship;
  • the plurality of first simulation workpieces are compared with the obtained distribution of the quality attribute values of the plurality of workpieces processed by the production equipment to determine the accuracy of the first relationship.
  • a workpiece data processing device which includes: at least one memory for storing computer-readable codes; at least one processor for executing the computer-readable codes stored in the at least one memory, and executing the first The method described in one aspect.
  • a method for setting processing parameters includes: generating a set of processing parameters of the production equipment under given ambient environmental condition data; for each element in the set, according to the setting Determine the quality attribute value of each of the plurality of second simulated workpieces processed according to the processing parameters represented by the element by determining the surrounding environment condition data, the element, and the first relationship, wherein the first relationship is the The relationship between the quality attribute value of the workpiece processed by the production equipment and the surrounding environmental condition data when the production equipment processes the workpiece and the processing parameters of the production equipment; for each element in the set, for a plurality of For each of the second simulated workpieces, the quality inspection result data of the second simulated workpiece is determined according to the second relationship and the determined quality attribute value of the second simulated workpiece, wherein the second relationship is determined State the relationship between the quality inspection result data and the quality attribute values of the workpieces processed by the production equipment; for each element in the set, statistical quality inspections are based on the quality inspection result data of each of the determined plural
  • a processing parameter setting device which includes: a processing parameter set establishment module configured to generate a set of processing parameters of the production equipment under given ambient environmental condition data; and an adjustment module , Configured to: for each element in the set, determine a plurality of second simulated workpieces to be processed according to the processing parameters represented by the element according to the given ambient environmental condition data, the element and the first relationship
  • the first relationship is the quality attribute value of the workpiece processed by the production equipment and the surrounding environmental condition data when the production equipment processes the workpiece and the processing parameters of the production equipment
  • the quality inspection result data wherein the second relationship is the relationship between the quality inspection result data of the workpiece processed by the production equipment and the quality attribute value; for each element in the set, according to the determined plural 2.
  • the quality inspection result data of each of the simulated workpieces is statistically quality inspection pass rate; the following process is repeated until the preset conditions are met, wherein the preset conditions include: the number of iterations reaches the maximum number of iterations or the calculated production equipment
  • the quality inspection pass rate of the processed workpiece reaches the pass rate threshold, and a new element is added to the set to make the expected value of the quality inspection pass rate obtained based on the new element statistics and the maximum value of the quality inspection pass rate obtained from previous statistics Compared with the largest increase; for the new element added to the set, according to the given ambient environmental condition data, the new element and the first relationship, determine the processing parameters represented by the new element
  • the optimal processing parameters under any surrounding environmental conditions can be automatically determined, so that the quality inspection pass rate of the workpiece is the highest.
  • the quality attribute values of the workpiece processed by the production equipment and the processing parameters of the production equipment are determined by processing the workpiece data , The first relationship between the surrounding environmental conditions when processing the workpiece, and the second relationship between the workpiece quality attribute value and the workpiece quality inspection result data.
  • a large number of workpiece quality attribute values can be simulated according to the determined first relationship, and then a large number of workpiece quality inspection result data can be simulated according to the determined second relationship, and the quality inspection pass rate can be counted accordingly. Because the simulated data meets statistical significance, and The simulation of data is based on the determined first relationship and second relationship, so the pass rate of quality inspection obtained by statistics can be considered relatively accurate. Furthermore, the optimal processing parameters can be obtained based on the relatively accurate quality inspection pass rate, and the optimal setting of the processing parameters of the production equipment is realized. On the one hand, by determining the first relationship and the second relationship, the relationship between the quality inspection results and the processing parameters of the production equipment can be more accurately described. On the other hand, according to the first relationship and the second relationship, a large amount of data is simulated, statistics are obtained to obtain a more accurate quality inspection pass rate, and the optimal processing parameter settings are obtained accordingly.
  • a Gaussian process can be fitted according to each element in the set and the pass rate of the element corresponding to the previous statistics, and then the Gaussian process can be used. The process calculates the new element.
  • a computer-readable medium stores computer-readable code, and when the computer-readable code is executed by at least one processor, the Methods.
  • Fig. 1 is a schematic diagram of an industrial system provided by an embodiment of the present invention.
  • Fig. 2 is a flowchart of a method for processing workpiece data according to an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of the process of determining two relationships in the workpiece data processing method provided by an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of using a CGAN model to determine the first relationship in the workpiece data processing method provided by an embodiment of the present invention.
  • Fig. 5 is a flowchart of a verification method provided by an embodiment of the present invention.
  • Fig. 6 is a flowchart of a method for setting processing parameters provided by an embodiment of the present invention.
  • FIG. 7 is a schematic diagram of comparing the simulated workpiece data generated by the CGAN model with the real workpiece data in the embodiment of the present invention.
  • Fig. 8 is a schematic structural diagram of a workpiece data processing device provided by an embodiment of the present invention.
  • FIG. 9 is a schematic diagram of another structure of a workpiece data processing device provided by an embodiment of the present invention.
  • Discriminator 702 Generator
  • 701c1, 701c2 the input vector of the discriminator 701
  • 702c1 Noise input vector of generator 702
  • 702c2 Input vector of generator 702
  • 702b Hidden layer of generator 702
  • Processing parameter setting method S601 Generate a collection of processing parameters
  • S602 Generate quality attribute values
  • S606 Generate quality for new elements
  • S607 Determine quality for new elements
  • S608 Statistic attribute values for new elements Test result data Quantity test pass rate
  • 301a Workpiece data acquisition module 301b: First relationship determination module 301c: Second relationship determination module
  • 303a Processing parameter set establishment module 303b: Adjustment module 303c: Optimal processing parameter determination module
  • a data-driven method is adopted to automatically set the processing parameters of the production equipment, and the optimal quality of the workpiece is realized by analyzing the workpiece data.
  • the optimal processing parameters under any surrounding environmental conditions can be automatically determined, so that the quality inspection pass rate of the workpiece is the highest.
  • the quality attribute values of the workpiece processed by the production equipment and the processing parameters of the production equipment are determined by processing the workpiece data , The first relationship between the surrounding environmental conditions when processing the workpiece, and the second relationship between the workpiece quality attribute value and the workpiece quality inspection result data.
  • a large number of workpiece quality attribute values can be simulated according to the determined first relationship, and then a large number of workpiece quality inspection result data can be simulated according to the determined second relationship, and the quality inspection pass rate can be counted accordingly.
  • the simulated data meets statistical significance, and The simulation of data is based on the determined first relationship and second relationship, so the pass rate of quality inspection obtained by statistics can be considered relatively accurate. Furthermore, the optimal processing parameters can be obtained based on the relatively accurate quality inspection pass rate, and the optimal setting of the processing parameters of the production equipment is realized. On the one hand, by determining the first relationship and the second relationship, the relationship between the quality inspection results and the processing parameters of the production equipment can be more accurately described. On the other hand, according to the first relationship and the second relationship, a large amount of data is simulated, statistics are obtained to obtain a more accurate quality inspection pass rate, and the optimal processing parameter settings are obtained accordingly.
  • the CGAN model can be used to obtain the relationship between the processing parameters of the production equipment and the distribution of the quality attributes of the workpiece under different ambient environmental conditions. For each combination of the processing parameters of the production equipment and the surrounding environmental conditions, only a small amount of data can be used to train the CGAN model. For any combination of processing parameters and surrounding environmental conditions, the trained CGAN model can generate any number of simulated workpiece samples. Using the method provided by the embodiment of the present invention can significantly reduce the data required to determine the relationship between the processing parameters and the result of the workpiece quality inspection.
  • KS test Kermo gorov-Smirnov test
  • KS test Kolmo gorov-Smirnov test
  • the simulated workpiece samples can be used for virtual quality inspection with almost no cost.
  • train a classifier that can predict the quality inspection result under the condition of a given workpiece quality attribute, and implement virtual quality inspection by training the classifier.
  • the Bayesian optimization algorithm can also be used to obtain virtual quality inspection results based on the simulated workpiece samples generated by the CGAN model and the simulated workpiece samples to automatically determine the optimal quality inspection result. Processing parameters.
  • Fig. 1 is a schematic diagram of an industrial system provided by an embodiment of the present invention.
  • the industrial system 100 includes a production equipment 10 and a plurality of workpieces 20 processed by the production equipment 10.
  • the industrial system 100 may further include a workpiece data processing device 30 configured to process data of the workpiece 20.
  • the workpiece data processing device 30 may include: a workpiece data processing device 301 configured to acquire data of the workpiece 20 and determine the quality attribute value of the workpiece processed by the production equipment 10 and the workpiece processed by the production equipment 10 through the data processing.
  • the relationship between the surrounding environmental condition data 90 and the quality data value of the workpiece processed by the production equipment 10 is called the “first relationship”; and the quality inspection result data 60 and the quality attribute value 50 of the workpiece processed by the production equipment 10 are determined
  • the relationship between is called the "second relationship”.
  • the workpiece data processing device 30 may further include a verification device 302 configured to generate data of a plurality of simulated workpieces according to the above-mentioned first relationship, and compare the generated data of the simulated workpiece with the data of the workpiece processed by the production equipment 10 To determine the accuracy of the first relationship.
  • a verification device 302 configured to generate data of a plurality of simulated workpieces according to the above-mentioned first relationship, and compare the generated data of the simulated workpiece with the data of the workpiece processed by the production equipment 10 To determine the accuracy of the first relationship.
  • the workpiece data processing device 30 may also include a processing parameter setting device 303 configured to determine the processing parameter 40 of the production equipment 10 under the given ambient environmental condition data 90 according to the above-mentioned first relationship and the second relationship. The optimal value.
  • Fig. 2 is a flowchart of a method for processing workpiece data according to an embodiment of the present invention.
  • the method 200 may be executed by the workpiece data processing device 30, as shown in FIG. 2, and may include the following steps:
  • step S201 the following data of each of the plurality of workpieces 20 processed by one production equipment 10 is acquired:
  • the processing condition data includes:
  • the processing parameters 40 used by the production equipment 10 when processing the workpiece 20 are the processing parameters 40 used by the production equipment 10 when processing the workpiece 20, and
  • Quality inspection result data 60 such as: passed or failed quality inspection.
  • the plurality of workpieces 20 may be part or all of the workpieces processed by the production equipment 10.
  • the acquired data try to cover various combinations of ambient environmental condition data 90 and processing parameters 40.
  • For each combination of ambient environmental condition data 90 and processing parameter 40 there is no need to limit the amount of acquired data. That is, in the embodiment of the present invention, there is no need to obtain a large amount of workpiece data to satisfy statistical significance, as long as the combination of various surrounding environmental condition data 90 and processing parameters 40 is covered as much as possible.
  • the quality inspection result data 60 of the workpiece processed by the production equipment 10 and the surrounding environmental condition data 90 of the production equipment 10 when the workpiece is processed and the processing parameters 40 of the production equipment 10 are not directly determined.
  • the reason is that if the relationship is directly determined, the amount of workpiece data required is relatively large in order to meet statistical significance.
  • the quality inspection result data 60 includes: passed quality inspection and failed quality inspection, then under the same combination of processing parameters 40 and quality inspection result data 60, due to the influence of some random factors, one may be processed by the production equipment 10
  • the quality attribute value 50 of the workpiece processed by the production equipment 10 and the surrounding area of the workpiece processed by the production equipment 10 are determined by methods such as generating a model 70.
  • the first relationship between the environmental condition data 90 and the processing parameter 40 of the production equipment 10, and then the second relationship between the quality inspection result data 60 and the quality attribute value 50 of the workpiece processed by the production equipment 10 is determined by a method such as the classifier 80 relationship. Since the quality attribute value 50 determines the quality inspection result data 60, the result is less affected by random factors. Therefore, the second relationship between the determined quality attribute value 50 and the quality inspection result data 60 is relatively accurate.
  • step S202 the quality attribute value 50 of the workpiece processed by the production equipment 10 and the surrounding environmental conditions of the production equipment 10 during the processing of the workpiece can be determined according to the processing condition data and the quality attribute value 50 of each of the workpieces 20.
  • the first relationship is determined by training the CGAN model 70.
  • the CGAN model 70 includes two sub-models: a generator 702 (Generator, "G” for short) and a discriminator 701 (Discriminator, "D” for short).
  • the generator 702 of the CGAN model 70 may include two input vectors (z, y) and one output vector ((G(z
  • the two input vectors include:
  • condition vector 702c2(y) is the processing condition data of each of the plural workpieces 20 (combination of the processing parameters 40 and the surrounding environment condition data 90)
  • y)) of the generator 702 is the quality attribute value 50 of the simulated workpiece.
  • the goal of the generator 702 is to generate a simulated quality attribute value of 50, so that the discriminator 701 cannot distinguish the simulated quality attribute value 50 from the real quality attribute value 50 (that is, the quality attribute value 50 of the workpiece 20 processed by the production equipment 10) open.
  • the goal can be expressed mathematically as:
  • p(z) is a mutually independent multivariate uniform distribution with a value of 0 to 1.
  • the discriminator 701 of the CGAN model 70 also has two input vectors and one output, and the output is a scalar.
  • the two input vectors include:
  • condition vector 701c2(y) is the same as a condition vector 702c2 of the generator 702.
  • the output of the discriminator 701 of the CGAN model 70 is a scalar (D(x'
  • the goal of the discriminator 701 is to distinguish between the real quality attribute value 50 and the simulated quality attribute value 50, which can be expressed mathematically as:
  • p(x) is the distribution of the true quality attribute value 50.
  • the first part of the above formula minimizes the output for the real quality attribute value 50; the second part maximizes the output for the simulated quality attribute value 50 output by the generator 702; the third part is the gradient penalty loss (see 2017 In 2005, Gulrajani, Ishaan, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C. Courville published an article "Improved Wasserstein GANs Training” on "Advances in Neural Information Processing Systems” on pages 5767 to 5777 ⁇ (Improved training of wasserstein gans), where ⁇ is a hyperparameter, usually taken as 10.
  • p(x') is uniform along the straight line between the real quality attribute value 50 and the simulated quality attribute value 50 Value to avoid the gradient disappearing or increasing sharply during training.
  • the CGAN model 70 can be trained by alternately minimizing the objective functions of the generator 702 and the discriminator 701 through multiple iterations until the discriminator 701 cannot divide the real quality attribute value 50 and the simulated quality attribute value 50.
  • the generator 702 can then be used to generate a large number of simulated quality attribute values 50 under any processing parameters 40 and any surrounding environmental condition data 90. Accordingly, the distribution of the simulated quality attribute value 50 is close to the distribution of the real quality attribute value 50.
  • FIG. 7 shows the distribution of the quality data value 50 of the actual workpiece collected from the worm gear manufacturer and the simulated quality attribute value 50 generated by the CGAN model under the same CNC machining parameters and surrounding environmental conditions.
  • the high-dimensional quality attribute values are mapped to two-dimensional for easy display through the dimensionality reduction method.
  • the solid circle represents the real quality attribute value of 50
  • the open circle represents the simulated quality attribute value of 50.
  • step S203 the second relationship between the quality inspection result data 60 and the quality attribute value 50 of the workpiece processed by the production equipment 10 can be determined according to the quality attribute value and quality inspection result data of each of the plurality of workpieces 20 .
  • the classifier 80 may be used to determine the second relationship. In fact, as long as the best classification accuracy can be achieved, you can choose any classifier, such as random forest, support vector machine, and even use the predefined acceptable range to check the quality attribute value 50 in this step.
  • the input of the classifier 80 is the quality attribute value 50 of the workpiece 20 obtained in step S201
  • the output is the quality inspection result data 60 of the workpiece 20.
  • the classifier 80 is trained according to the real data obtained in step S201. After the training, the classifier 80 can be used to classify whether the simulated quality attribute value 50 passes the quality test, which can be regarded as a virtual quality test.
  • Fig. 5 is a flowchart of a verification method provided by an embodiment of the present invention.
  • the verification method 500 can be executed by the verification device 302 in FIG. 1. As shown in Figure 5, the following steps can be included:
  • step S501 the quality attribute values 50 of the plurality of first simulated workpieces 20 can be generated according to the processing condition data 40, the quality attribute value 50, and the first relationship of each of the plurality of workpieces 20 processed by the production equipment 10 .
  • the generator 702 in the CGAN model 70 can be used here to generate a large number of simulated quality attribute values 50.
  • step S502 Compare the first simulated workpiece 20 generated in step S501 with the acquired distribution of the quality attribute values of the plurality of workpieces 20 processed by the production equipment 10 to determine the accuracy of the first relationship.
  • the Kolmogorov-Smirnov test can be used to quantitatively measure the similarity between the real quality attribute value 50 and the simulated quality attribute value 50 to ensure that the simulated quality attribute value 50 can fully represent the real quality attribute value 50.
  • the Kolmogorov-Smirnov test (k-s test or k s test) is a non-parametric test that can be used to compare whether the continuous one-dimensional probability distributions of two samples are equal.
  • the multivariate generalization of the test is used to check whether the difference between the simulated quality attribute value 50 and the real quality attribute value 50 is statistically significant (see Jerome H. Friedman and Lawrence C.
  • the quality inspection pass rate of the workpiece can be evaluated under any ambient environmental conditions and under arbitrarily set processing parameters.
  • a brute force method for seeking the optimal setting of processing parameters of production equipment is to enumerate all possible parameter settings and select the parameter with the highest passing rate of quality inspection.
  • this method is too expensive and sometimes unacceptable, especially when the surrounding environmental conditions change rapidly and the optimal processing parameters need to be dynamically adjusted. Therefore, the embodiment of the present invention also provides a processing parameter setting method, as shown in FIG. 6.
  • Fig. 6 is a flowchart of a method for setting processing parameters provided by an embodiment of the present invention.
  • the processing parameter setting method 600 can be executed by the processing parameter setting device 303 in FIG. 1. As shown in Figure 6, the method may include the following steps:
  • S601 Generate a set of processing parameters 40.
  • step S601 under given ambient environmental condition data, a set of processing parameters 40 of the production equipment 10 is generated, wherein the set of processing parameters 40 can be randomly generated.
  • the set of processing parameters 40 can be randomly generated.
  • M machining parameters 40 are randomly generated, where M is a positive integer.
  • step S602 for each element in the set generated in step S601, the quality attribute value 50 of a plurality of second simulated workpieces 20 is generated according to the given surrounding environment condition data 90, the element and the first relationship. Still taking CNC as an example, given the surrounding environment condition data 90, for each of the M processing parameters 40 randomly generated in the foregoing step S601, a large number of quality attribute values 50 are respectively generated using the CGAN model 70.
  • step S603 for each element in the set, the quality inspection result data 60 of each second simulated workpiece 20 is determined according to the second relationship and the generated quality attribute value 50 of the plurality of second simulated workpieces 20. Still taking CNC as an example, for each of the M processing parameters 40 randomly generated in the foregoing step S601, using the foregoing second relationship, according to a large number of quality attribute values 50 corresponding to the processing parameter 40 generated in the step S602 , Generate quality inspection result data 60 corresponding to each quality attribute value 50.
  • step S604 for each element in the set, the quality inspection pass rate corresponding to the element is calculated according to the determined quality inspection result data 60 of the plurality of second simulated workpieces 20 corresponding to the element.
  • the machining parameter 40 is used for statistics, and the workpiece processed by the CNC The pass rate of quality inspection.
  • step S605 to step S608 are repeated until the preset condition is satisfied.
  • the preset condition can be set according to the actual situation. For example: the number of iterations reaches the maximum number of iterations; another example: the calculated quality inspection pass rate of the workpiece processed by the production equipment 10 reaches the pass rate threshold; another example: the number of iterations reaches the maximum number of iterations or the calculated workpiece processed by the production equipment 10 If one of these two conditions is met, the pass rate of the quality inspection reaches the pass rate threshold, and the cycle is exited.
  • a new element is added to the set, so that the expected value of the pass rate of quality inspection obtained by the statistics of the new element has the largest increase compared with the maximum value of the pass rate of quality inspection obtained by previous statistics.
  • a Gaussian Process GP
  • GP Gaussian Process
  • All elements in the set, that is, each processing parameter 40, are used as the input of the Gaussian process, and the quality inspection pass rate corresponding to each processing parameter 40 is used as the output of the Gaussian process. .
  • Bayesian optimization can be used to effectively search for processing parameters 40 Candidate space. Then, under 40 settings of each processing parameter in the given set, the Gaussian process is used to estimate the posterior distribution of the pass rate of the quality inspection of all workpieces that have not been searched for settings:
  • x is the unsearched setting
  • ⁇ is the parameter of the Gaussian process
  • D is the estimated value of each search setting and the corresponding workpiece quality inspection pass rate.
  • the optimal configuration of the processing parameters 40 is realized under the specific ambient environment condition data 90.
  • step S606 the quality of each of the plurality of third simulated workpieces 20 produced according to the processing parameters represented by the new element is determined according to the given ambient environmental condition data, the new element, and the above-mentioned first relationship. Property 50.
  • step S607 for the new element, for each of the plurality of third simulated workpieces 20, the plurality of third simulated workpieces 20 are determined according to the above-mentioned second relationship and the determined quality attribute values of the plurality of third simulated workpieces.
  • the quality inspection result data 60 of each of the third simulated workpieces 20 is determined according to the above-mentioned second relationship and the determined quality attribute values of the plurality of third simulated workpieces.
  • step S608 for the new element added to the set, the quality inspection pass rate is calculated according to the determined quality inspection result data of each of the plurality of third simulation workpieces 20.
  • S609 Determine whether the above preset conditions are met. If the above-mentioned preset conditions are met, then jump out of the loop and execute step S610; otherwise, return to step S605.
  • the element corresponding to the maximum value of all the quality inspection pass rates obtained by statistics is determined as the optimal value of the processing parameter of the production equipment 10 under the given ambient environmental condition data.
  • the number of iterations is N, and the initial setting is 0.
  • the values of N and the number of elements M in the set of processing parameters 40 generated in step S601 may be determined by the number of processing parameters 40 that need to be considered. Taking CNC as an example, the number of CNC machining parameters is usually less than 10, so here you can consider setting M to 5 and N to 10. Specifically, the values of M and N can be adjusted according to experiments.
  • Fig. 8 is a schematic structural diagram of a workpiece data processing device provided by an embodiment of the present invention.
  • the workpiece data processing device 30 may include: a workpiece data processing device 301, and optionally, may also include a verification device 302 and/or a processing parameter setting device 303.
  • the workpiece data processing device 301 may include: a workpiece data acquisition module 301a configured to acquire processing condition data, quality attribute values, and quality inspection results of each of the plurality of workpieces 20 processed by one production equipment 10 Data
  • the processing condition data of a workpiece 20 includes: the processing parameters used by the production equipment 10 when processing the workpiece 20 and the surrounding environment condition data when the production equipment 10 processes the workpiece 20; a first relationship determination module 301b, It is configured to determine the quality attribute value of the workpiece processed by the production equipment 10 and the surrounding environmental condition data and the production equipment 10 when the production equipment 10 processes the workpiece according to the processing condition data and quality attribute value of each of the workpieces 20
  • a second relationship determination module 301c is configured to determine the quality of the workpiece processed by the production equipment 10 according to the quality attribute value of each workpiece 20 in the plurality of workpieces 20 and the quality inspection result data
  • the workpiece data acquisition module 301a is specifically configured to acquire processing condition data, quality attribute values, and quality inspection result data of each of the plurality of workpieces 20 processed by one production equipment 10, so that the acquired All data try to cover the combination of various surrounding environmental condition data and processing parameters.
  • the first relationship determination module 301b is specifically configured to: use the processing condition data of each workpiece 20 in the plurality of workpieces 20 as an input vector of the generator 702 in a CGAN model;
  • the processing condition data of each workpiece 20 is used as an input vector of the discriminator 701 of the CGAN model;
  • the quality attribute value of each of the plurality of workpieces 20 or the output vector 702a of the generator of the CGAN model is used as the CGAN model
  • Another input vector of the discriminator 701; the CGAN model is trained, and the generator 701 of the trained CGAN model is used as the first relationship.
  • the verification device 302 may include: a simulated workpiece data generation module 302a, configured to generate a complex number according to the processing condition data, quality attribute value, and first relationship of each of the plurality of workpieces 20 processed by the production equipment 10 The quality attribute values of the first simulated workpiece 20; a data comparison module 302b configured to compare the first simulated workpieces 20 with the acquired distribution of the quality attribute values of the plurality of workpieces 20 processed by the production equipment 10, To determine the accuracy of the first relationship.
  • a simulated workpiece data generation module 302a configured to generate a complex number according to the processing condition data, quality attribute value, and first relationship of each of the plurality of workpieces 20 processed by the production equipment 10 The quality attribute values of the first simulated workpiece 20
  • a data comparison module 302b configured to compare the first simulated workpieces 20 with the acquired distribution of the quality attribute values of the plurality of workpieces 20 processed by the production equipment 10, To determine the accuracy of the first relationship.
  • the processing parameter setting device 303 may include:
  • a processing parameter set establishment module 303a is configured to generate a set of processing parameters of the production equipment 10 under given ambient environmental condition data
  • An adjustment module 303b is configured to, for each element in the set, determine a plurality of second simulation workpieces 20 to be processed according to the processing parameters represented by the element according to the given ambient environmental condition data, the element and the first relationship The quality attribute value of each of them, where the first relationship is the relationship between the quality attribute value of the workpiece processed by the production equipment 10 and the surrounding environmental condition data when the production equipment 10 processes the workpiece and the processing parameters of the production equipment 10; For each element in the set, for each of the plurality of second simulated workpieces 20, the quality inspection result data of the second simulated workpiece 20 is determined according to the second relationship and the determined quality attribute value of the second simulated workpiece 20 , Where the second relationship is the relationship between the quality inspection result data of the workpiece processed by the production equipment 10 and the quality attribute value; for each element in the set, the quality of each of the plurality of second simulated workpieces 20 is determined Statistical quality inspection pass rate of inspection result data;
  • the adjustment module 303b is further configured to repeat the following process until a preset condition is met, where the preset condition includes: the number of iterations reaches the maximum number of iterations or the calculated quality inspection pass rate of the workpiece processed by the production equipment 10 reaches the pass rate threshold: A new element is added to the set, so that the expected value of the pass rate of quality inspection based on the statistics of the new element has the largest increase compared to the maximum value of the pass rate of quality inspection obtained from the previous statistics; for the new element added to the set , According to the given ambient environmental condition data, the new element and the first relationship, determine the quality attribute value of each of the plurality of third simulation workpieces 20 produced according to the processing parameters represented by the new element; For each of the plurality of third simulation workpieces 20, according to the second relationship and the determined quality attribute value of the plurality of third simulation workpieces 20, each of the plurality of third simulation workpieces 20 One piece of quality inspection result data; for the new element added to the set, the quality inspection pass rate is calculated according to the determined
  • An optimal processing parameter determination module 303c is configured to determine the element corresponding to the maximum value of all the quality inspection pass rates obtained by statistics, as the optimal value of the processing parameter of the production equipment 10 under the given ambient environmental condition data.
  • the adjustment module 303b adds a new element to the set, so that the expected value of the pass rate of quality inspection obtained based on the new element statistics has the largest increase compared to the maximum value of the pass rate of quality inspection obtained by previous statistics. It is configured to fit a Gaussian process according to each element in the set and the passing rate of the quality inspection corresponding to the element obtained from previous statistics; and calculate the new element by using the Gaussian process.
  • FIG. 9 is a schematic diagram of another structure of a workpiece data processing device 30 provided by an embodiment of the present invention.
  • the workpiece data processing device 30 may include at least one memory 304 for storing computer-readable codes; at least one processor 305 is configured to execute the computer-readable information stored in the at least one memory 304
  • the code executes the aforementioned method 200, and optionally, the aforementioned method 500 and/or method 600 can also be executed.
  • the modules shown in FIG. 8 can be regarded as program modules written by computer readable codes stored in the memory 304, and when these program modules are called by the processor 305, the aforementioned methods 200, 500, and/or 600 can be executed.
  • the workpiece data processing device 30 may also include an I/O interface 306, which can be connected to external devices such as a mouse and a display.
  • an I/O interface 306 can be connected to external devices such as a mouse and a display.
  • the at least one memory 304, the at least one processor 305, and the I/O interface 306 can communicate through the bus 307.
  • an embodiment of the present invention also provides a computer-readable medium that stores computer-readable code, and when the computer-readable code is executed by at least one processor, the foregoing methods 200, 500 and/or are implemented. 600.
  • the embodiments of the present invention provide a workpiece data processing method, device, and computer readable medium, which are used to accurately determine the relationship between the processing parameters of the production equipment, the ambient environmental condition data, and the quality inspection result of the workpiece.
  • the optimal processing parameters can be determined based on the relationship.
  • the workpiece data processing method provided by the embodiment of the present invention is purely data-driven. Compared with the method driven by human experience, the accuracy, efficiency and versatility of the process of setting optimal processing parameters for workpiece processing are significantly improved.
  • the solution provided by the embodiment of the present invention is cost-effective and predicts the throughput rate of the workpiece under given processing parameters and ambient environmental conditions to construct the data volume required for the mathematical model.
  • the solution provided by the embodiment of the present invention only needs to search a small part of the candidate space of processing parameters to find the best processing parameters.
  • the CGAN model is used to simulate the quality attribute value of the workpiece under given processing parameters, and then the classifier is used to perform virtual quality inspection to determine the quality inspection pass rate of the workpiece.
  • the Bayesian optimization method can be used to automatically and effectively find the best setting of the processing parameters of the production equipment, so as to obtain the highest pass rate of workpiece quality inspection under any surrounding environmental conditions.
  • system structure described in the foregoing embodiments may be a physical structure or a logical structure. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by at least two physical entities, or at least Some components in two independent devices are implemented together.
  • the hardware unit can be implemented mechanically or electrically.
  • a hardware unit may include permanent dedicated circuits or logic (such as dedicated processors, Field-Programmable Gate Array (FPGA) or Application Specific Integrated Circuits (ASIC), etc.). Complete the corresponding operation.
  • the hardware unit may also include programmable logic or circuits (such as general-purpose processors or other programmable processors), which may be temporarily set by software to complete corresponding operations.
  • the specific implementation mode mechanical method, or dedicated permanent circuit, or temporarily set circuit

Landscapes

  • Engineering & Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Human Computer Interaction (AREA)
  • General Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • General Factory Administration (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

涉及工业大数据技术领域，尤其涉及一种工件数据处理方法、装置和计算机可读介质，用于准确确定生产设备的加工参数、周围环境条件数据与工件的质量检验结果之间的关系，进而能够依据该关系确定最优的加工参数。一种工件数据方法包括：获取由一个生产设备所加工的复数个工件中每一个工件的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件的加工条件数据包括：所述生产设备在加工该工件时所使用的加工参数以及所述生产设备在加工该工件时的周围环境条件数据；确定所述生产设备所加工工件的质量属性值与所述生产设备在加工工件时的周围环境条件数据和所述生产设备的加工参数之间的第一关系；以及确定所述生产设备所加工工件的质量检验结果数据与质量属性值之间的第二关系。

Description

生产设备的加工参数设置方法、装置和计算机可读介质 技术领域
本发明涉及工业大数据技术，尤其涉及一种工件数据处理方法、装置和计算机可读介质。
背景技术
一个生产设备所加工工件的质量很大程度取决于该生产设备的参数设置。以数控机床(CNC)为例，CNC加工参数的设置，比如：切割位置、切割角度、轴高度和装夹位置等的变化都会导致所加工工件质量属性值的变化。一个质量过关的工件必须要经过质量检验，即其质量属性值在质量检验中需要过关(通常，不同类型的工件需要检验不同的质量属性，比如：对于汽车工业中的蜗轮(worm gear)，需要检验平均扭矩、传动比、峰值转速等)。因此，生产设备加工参数的设置对于工件制造，特别是精加工行业的工件制造过程中工件质量的优化至关重要。
目前，诸如CNC等生产设备的加工参数通常依靠人的经验来设置，而参数的最优设置可能会受到工件的各种质量属性可接受范围的影响，因此，仅依靠人的经验往往很难实现加工参数的最优设置。此外，工件质量容易受到周围环境条件影响，比如：工件制造过程中温度和湿度的影响。因此，在设置加工参数时还要考虑到周围环境条件的变化，这更增大了加工参数设置的难度。
如前所述，工件质量与生产设备加工参数的设置相关，而现有仅依靠人的经验进行加工参数设置很难使得工件质量最优。而工件质量往往通过质量检验来衡量，因此如何设置生产设备加工参数以提高工件质量检验通过率是一个亟待解决的问题。
一种可能的方法是，通过确定生产设备加工参数与工件质量检验通过率之间的关系，实现生产设备加工参数的自动最优设置。可以通过机器学习的算法来确定工件生产设备的加工参数与工件质量检验的通过率之间的关系。一种方式是采用监督学习的算法，通过对大量已标注数据进行学习来建立模型，确定关系。但对于制造工业而言，数据收集和标注往往比较困难且耗资巨大，这种情况在未数据化的工厂中尤为突出。若想确定一个生产设备在制造过程中的加工参数和其所加工工件的质量检验通过率之间的关系往往需要大量数据。假定欲确定在下列CNC加工参数设置下的蜗轮的质量通过率：y轴切割位置为91，螺旋角为-2.97，z轴高度为209.8，周围环境温度为7摄氏度，湿度为70％。完成上述处理通常需要收集几百个 蜗轮样本以保证统计显著性，从而能够计算得到在CNC加工参数和周围环境条件的单一组合下准确的蜗轮质量检验通过率。而在生产实践中，通常无法得到足够的数据，对于每一种CNC加工参数和周围环境条件的组合往往仅有少数几项标注的数据，借助传统的回归模型难以获取在不同的周围环境条件下CNC加工参数和工件质量检验结果的准确关系。
发明内容
本发明实施例提供了一种工件数据处理方法、装置和计算机可读介质，用于准确确定生产设备的加工参数、周围环境条件数据与工件的质量检验结果之间的关系。进而能够依据该关系确定最优的加工参数。
第一方面，提供一种工件数据处理方法，包括：获取由一个生产设备所加工的复数个工件中每一个的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件的加工条件数据包括：所述生产设备在加工该工件时所使用的加工参数以及所述生产设备在加工该工件时的周围环境条件数据；根据所述复数个工件中每一个的加工条件数据和质量属性值，确定所述生产设备所加工工件的质量属性值与所述生产设备在加工工件时的周围环境条件数据和所述生产设备的加工参数之间的第一关系；根据所述复数个工件中每一个的质量属性值以及质量检验结果数据，确定所述生产设备所加工工件的质量检验结果数据与质量属性值之间的第二关系。
第二方面，提供一种工件数据处理装置，包括：一个工件数据获取模块，被配置为获取由一个生产设备所加工的复数个工件中每一个的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件的加工条件数据包括：所述生产设备在加工该工件时所使用的加工参数以及所述生产设备在加工该工件时的周围环境条件数据；一个第一关系确定模块，被配置为根据所述复数个工件中每一个的加工条件数据和质量属性值，确定所述生产设备所加工工件的质量属性值与所述生产设备在加工工件时的周围环境条件数据和所述生产设备的加工参数之间的第一关系；一个第二关系确定模块，被配置为根据所述复数个工件中每一个的质量属性值以及质量检验结果数据，确定所述生产设备所加工工件的质量检验结果数据与质量属性值之间的第二关系。
和以往直接确定生产设备加工参数与工件质量检验通过率之间关系不同的是，本发明实施例中，通过对工件数据进行处理，确定生产设备所加工工件的质量属性值与生产设备的加 工参数、加工工件时的周围环境条件之间的第一关系，以及工件质量属性值与工件质量检验结果数据之间的第二关系。通过确定第一关系和第二关系，能够较准确地描述质量检验结果与生产设备的加工参数之间的关系。
可选地，在获取由一个生产设备所加工的复数个工件中每一个的加工条件数据、质量属性值以及质量检验结果数据时，可获取由所述生产设备所加工的复数个工件中每一个的加工条件数据、质量属性值以及质量检验结果数据，以使得获取的所有数据尽量覆盖各种周围环境条件数据和加工参数的组合。这样，对于生产设备的加工参数与周围环境条件的每一种组合，可仅使用少量数据来训练得到第一关系的数学模型。
可选地，可使用条件生成对抗网络(Conditional Generative Adversarial Net，CGAN)模型来获取不同周围环境条件下生产设备加工参数与工件质量属性的分布之间的第一关系。具体地，可将所述复数个工件中每一个的加工条件数据作为一个CGAN模型中生成器的一个输入向量；将所述复数个工件中的每一个的加工条件数据作为所述CGAN模型的判别器的一个输入向量；将所述复数个工件中的每一个工件的质量属性值或所述CGAN模型的生成器的输出向量，作为所述CGAN模型中判别器的另一个输入向量；训练所述CGAN模型，将训练得到的所述CGAN模型的生成器作为所述第一关系。对于生产设备的加工参数与周围环境条件的每一种组合，可仅使用少量数据来训练得到CGAN模型。对于任一种加工参数和周围环境条件的组合，训练得到的CGAN模型可生成任意数量的模拟的工件样本。采用本发明实施例提供的方法，可显著减少确定加工参数和工件质量检验结果之间的关系所需的数据。
可选地，可根据所述生产设备所加工的复数个工件中每一个的加工条件数据、质量属性值，以及所述第一关系生成复数个第一模拟工件中每一个的质量属性值；将所述复数个第一模拟工件与获取的所述生产设备所加工的所述复数个工件的质量属性值的分布相比较，以确定所述第一关系的准确性。通过度量真实的工件样本和模拟的工件样本之间的相似程度，可确保模拟的样本能够充分代表真实的工件。
第三方面，提供一种工件数据处理装置，包括：至少一个存储器，用于存放计算机可读代码；至少一个处理器，用于执行所述至少一个存储器存储的所述计算机可读代码，执行第一方面所述的方法。
第四方面，提供一种加工参数设置方法，包括：在给定的周围环境条件数据下，生成一个所述生产设备的加工参数的集合；对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及所述第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件中每一个的质量属性值，其中，所述第一关系为所述生产设备所加工工件的质量属性值与所述生产设备在加工工件时的周围环境条件数据和所述生产设备的加工参数之间的关系；对于所述集合中的每一个元素，针对复数个第二模拟工件中的每一个，根据所述第二关系，以及确定的该第二模拟工件的质量属性值，确定该第二模拟工件的质量检验结果数据，其中，所述第二关系为所述生产设备所加工工件的质量检验结果数据与质量属性值之间的关系；对于所述集合中的每一个元素，根据确定的复数个第二模拟工件中每一个的质量检验结果数据统计质量检验通过率；重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备所加工工件的质量检验通过率达到通过率阈值，在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所加工的复数个第三模拟工件中每一个的质量属性值；对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件中的每一个，根据所述第二关系，以及确定的所述复数个第三模拟工件的质量属性值，确定所述复数个第三模拟工件中每一个的质量检验结果数据；对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件中每一个的质量检验结果数据统计质量检验通过率；确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备的加工参数的最优值。
第五方面，提供了一种加工参数设置装置，包括：一个加工参数集合建立模块，被配置为在给定的周围环境条件数据下，生成一个所述生产设备的加工参数的集合；一个调整模块，被配置为：对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件中每一个的质量属性值，其中，所述第一关系为所述生产设备所加工工件的质量属性值与所述生产设备在加工工件时的周围环境条件数据和所述生产设备的加工参数之间的关系；对于所述集合中的每一个元素，针对复数个第二模拟工件中的每一个，根据第二关系以及确定的该第二模拟工件的质量属性值，确定该第二模拟工件的质量检验结果数据，其中，所述第二关系为所述生产设 备所加工工件的质量检验结果数据与质量属性值之间的关系；对于所述集合中的每一个元素，根据确定的复数个第二模拟工件中每一个的质量检验结果数据统计质量检验通过率；重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备所加工工件的质量检验通过率达到通过率阈值，在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所生产的复数个第三模拟工件中每一个的质量属性值；对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件中的每一个，根据所述第二关系以及确定的所述复数个第三模拟工件的质量属性值，确定所述复数个第三模拟工件中每一个的质量检验结果数据；对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件中每一个的质量检验结果数据统计质量检验通过率；一个最优加工参数确定模块，确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备的加工参数的最优值。
采用本发明实施例，可自动确定在任意周围环境条件下最优的加工参数，以使得工件的质量检验通过率最高。和以往直接确定生产设备加工参数与工件质量检验通过率之间关系不同的是，本发明实施例中，通过对工件数据进行处理，确定生产设备所加工工件的质量属性值与生产设备的加工参数、加工工件时的周围环境条件之间的第一关系，以及工件质量属性值与工件质量检验结果数据之间的第二关系。可根据确定的第一关系模拟大量的工件质量属性值，再根据确定的第二关系模拟大量的工件质量检验结果数据，并据此统计质量检验通过率，由于模拟的数据满足统计显著性，且数据的模拟是依据已确定的第一关系和第二关系，因此统计得到的质量检验通过率可认为是相对准确的。进而再依据相对准确的质量检验通过率可得到最优的加工参数，实现了生产设备加工参数的最优设置。一方面，通过确定第一关系和第二关系，能够较准确地描述质量检验结果与生产设备的加工参数之间的关系。另一方面，根据第一关系和第二关系，模拟大量数据，统计得到较准确的质量检验通过率，并据此得到最优加工参数的设置。
可选地，在所述集合中加入一个新的元素时，可根据所述集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程，进而利用所述高斯过程计算得到所述新的元素。
第六方面，提供一种计算机可读介质，所述计算机可读介质存储有计算机可读代码，当所述计算机可读代码被至少一个处理器执行时，执行第一方面或第四方面所提供的方法。
附图说明
图1为本发明实施例提供的工业系统的示意图。
图2为本发明实施例提供的工件数据处理方法的流程图。
图3为本发明实施例提供的工件数据处理方法中两种关系的确定过程示意图。
图4为本发明实施例提供的工件数据处理方法中，采用CGAN模型确定第一关系的示意图。
图5为本发明实施例提供的验证方法的流程图。
图6为本发明实施例提供的加工参数设置方法的流程图。
图7为本发明实施例中采用CGAN模型生成模拟工件数据与真实工件数据比较的示意图。
图8为本发明实施例提供的工件数据处理装置的结构示意图。
图9为本发明实施例提供的工件数据处理装置的又一结构示意图。
附图标记列表：
100：工业系统                10：生产设备                20：生产设备10所加工工件
30：工件数据处理装置
301：工件数据处理装置        302：验证装置               303：加工参数设置装置
200：工件数据处理方法
S201：获取工件数据           S202：确定第一关系          S203：确定第二关系
40：加工参数                 50：质量属性值              60：质量检验结果
70：生成模型                 80：分类器                  90：周围环境条件数据
701：判别器                  702：生成器
701c1，701c2：判别器701的输入向量
701b：判别器的隐藏层         701a：判别器701的输出
702c1：生成器702的噪声输入向量 702c2：生成器702的输入向量  702b：生成器702的隐藏层
702a：生成器702的输出向量
500：验证方法                S501：生成工件数据          S502：数据比较
600：加工参数设置方法        S601：生成加工参数的集合    S602：生成质量属性值
S603：确定质量检验结果数据  S604：统计质量检验通过率    S605：在集合中加入新的元素
S606：针对新的元素生成质量  S607：针对新的元素确定质量  S608：针对新的元素统计质属性值                     检验结果数据              量检验通过率
S609：判断是否满足预设条件  S610：确定加工参数的最优值
301a：工件数据获取模块      301b：第一关系确定模块      301c：第二关系确定模块
302a：模拟工件数据生成模块  302b：数据比较模块
303a：加工参数集合建立模块  303b：调整模块              303c：最优加工参数确定模块
304：存储器                 305：处理器                 306：I/O(输入/输出)接口
307：总线
具体实施方式
本发明实施例中，采用了数据驱动的方法来自动设置生产设备的加工参数，通过对工件数据的分析来实现工件质量最优。采用本发明实施例，可自动确定在任意周围环境条件下最优的加工参数，以使得工件的质量检验通过率最高。
和以往直接确定生产设备加工参数与工件质量检验通过率之间关系不同的是，本发明实施例中，通过对工件数据进行处理，确定生产设备所加工工件的质量属性值与生产设备的加工参数、加工工件时的周围环境条件之间的第一关系，以及工件质量属性值与工件质量检验结果数据之间的第二关系。可根据确定的第一关系模拟大量的工件质量属性值，再根据确定的第二关系模拟大量的工件质量检验结果数据，并据此统计质量检验通过率，由于模拟的数据满足统计显著性，且数据的模拟是依据已确定的第一关系和第二关系，因此统计得到的质量检验通过率可认为是相对准确的。进而再依据相对准确的质量检验通过率可得到最优的加工参数，实现了生产设备加工参数的最优设置。一方面，通过确定第一关系和第二关系，能够较准确地描述质量检验结果与生产设备的加工参数之间的关系。另一方面，根据第一关系和第二关系，模拟大量数据，统计得到较准确的质量检验通过率，并据此得到最优加工参数的设置。
本发明一些实施例中，可使用CGAN模型来获取不同周围环境条件下生产设备加工参数与工件质量属性的分布之间的关系。对于生产设备的加工参数与周围环境条件的每一种组合，可仅使用少量数据来训练得到CGAN模型。对于任一种加工参数和周围环境条件的组合，训练得到的CGAN模型可生成任意数量的模拟的工件样本。采用本发明实施例提供的方法，可显著减少确定加工参数和工件质量检验结果之间的关系所需的数据。
此外，本发明一些实施例中，还可使用KS检验(Kolmo gorov-Smirnov test)来度量真实的工件样本和模拟的工件样本之间的相似程度，以确保模拟的样本能够充分代表真实的工件。
模拟的工件样本可用于虚拟的质量检验，且几乎不需要成本。特别地，训练一个分类器，该分类器可在给定的工件质量属性的条件下预测质量检验结果，并通过训练该分类器实现虚拟的质量检验。
在一些实施例中，还可使用贝叶斯优化算法，基于CGAN模型产生的模拟的工件样本以及针对这些模拟的工件样本得到虚拟的质量检验结果，来自动确定保证最优质量检验结果的最优加工参数。
为了使本发明实施例的目的、技术方案和优点更加清楚明白，以下参照附图对本发明实施例进一步详细说明。其中，后续描述的实施例仅仅是本发明实施例的一部分，而非全部的实施例。
图1为本发明实施例提供的工业系统的示意图。如图1所示，该工业系统100包括一个生产设备10和该生产设备10所加工的复数个工件20。此外，该工业系统100还可包括一个工件数据处理装置30，被配置为处理工件20的数据。
其中，工件数据处理装置30可包括：一个工件数据处理装置301，被配置为获取工件20的数据，并通过对数据的处理确定生产设备10所加工工件的质量属性值与生产设备10在加工工件时的周围环境条件数据90以及生产设备10所加工工件的质量数据值之间的关系，称作“第一关系”；并且确定生产设备10所加工工件的质量检验结果数据60与质量属性值50之间的关系，称作“第二关系”。
工件数据处理装置30还可包括一个验证装置302，被配置为根据上述第一关系生成复数个模拟工件的数据，并将生成的模拟工件的数据与了生产设备10所加工的工件的数据进行比较，以确定第一关系的准确性。
此外，工件数据处理装置30还可包括一个加工参数设置装置303，被配置为依据上述第一关系和第二关系，确定在给定的周围环境条件数据90下，生产设备10的加工参数40的最优值。
图2为本发明实施例提供的工件数据处理方法的流程图。该方法200可由工件数据处理装置30执行，如图2所示，可包括如下步骤：
S201：获取工件数据。
在步骤S201中，获取由一个生产设备10所加工的复数个工件20中每一个工件20的下列数据：
1)加工条件数据
其中，加工条件数据包括：
生产设备10在加工该工件20时所使用的加工参数40，以及
生产设备10在加工该工件20时的周围环境条件数据90；
2)质量属性值50
3)质量检验结果数据60，比如：通过质量检验或未通过质量检验。
这里，复数个工件20可以是生产设备10所加工的工件中的部分或全部。在步骤S201，获取的数据尽量覆盖各种周围环境条件数据90和加工参数40的组合。对于每一种周围环境条件数据90和加工参数40的组合，可无需限定获取的数据量。即本发明实施例中，无需要求获取大量的工件数据以满足统计显著性，只要尽量覆盖各种周围环境条件数据90和加工参数40的组合即可。
在步骤S201之后，本发明实施例中，并没有直接确定生产设备10所加工工件的质量检验结果数据60与生产设备10在加工工件时的周围环境条件数据90和生产设备10的加工参数40之间的关系，这是因为，若直接确定该关系，为满足统计显著性，需要的工件数据量比较大。假设质量检验结果数据60包括：通过质量检验和未通过质量检验，那么在相同的加工参数40和质量检验结果数据60的组合下，由于一些随机因素的影响，可能由生产设备10加工得到的一个工件20通过质量检验，而另一个工件20却未通过质量检验。如果数据量较少，即工件样本较少的情况下，统计得到的关系并不准确。如图3所示，本发明实施例中，在工件数据较少的情况下，通过诸如生成模型70等方式确定生产设备10所加工工件的质量属性值50与生产设备10在加工工件时的周围环境条件数据90和生产设备10的加工参数40之间的第一关系，再通过诸如分类器80的方式确定生产设备10所加工工件的质量检验结果数据60与质量属性值50之间的第二关系。由于由质量属性值50来确定质量检验结果数据60，结果受随机因素的影响较小，因此，确定的质量属性值50与质量检验结果数据60之间的第二关系相对准确。
S202：确定第一关系。
在步骤S202中，可根据复数个工件20中每一个工件20的加工条件数据和质量属性值50，确定生产设备10所加工工件的质量属性值50与生产设备10在加工工件时的周围环境条件数 据90和生产设备10的加工参数40之间的第一关系。
可选地，通过训练CGAN模型70来确定第一关系。具体地，如图4所示，CGAN模型70包括两个子模型：即生成器702(Generator，简称“G”)和判别器701(Discriminator，简称“D”)。
CGAN模型70的生成器702可包括两个输入向量(z，y)和一个输出向量((G(z|y)))。
其中，两个输入向量包括：
1)一个噪声向量702c1(z)，包括随机生成的取值0～1之间的浮点数。
2)条件向量702c2(y)，是复数个工件20中每一个工件20的加工条件数据(加工参数40和周围环境条件数据90的组合)
生成器702的输出向量(G(z|y))为模拟的工件的质量属性值50。
生成器702的目标是生成模拟的质量属性值50，使得判别器701也无法将模拟的质量属性值50和真实的质量属性值50(即生产设备10所加工工件20的质量属性值50)区分开。该目标在数学上可表示为：
Figure PCTCN2019085055-appb-000001
其中，p(z)为取值为0到1的相互独立的多变量均匀分布。
CGAN模型70的判别器701也有两个输入向量和一个输出，该输出为一个标量。
其中，两个输入向量包括：
1)条件向量701c2(y)，与生成器702的一个条件向量702c2相同。
2)模拟的工件的质量属性值50(G(z|y))或真实的工件的质量属性值50(即步骤S201中获取的生产设备10所加工工件的质量属性值50)。
CGAN模型70的判别器701的输出为一个标量(D(x’|y))，该值越大表示判别器701认为x’是模拟工件的质量属性值50的可能性越大。判别器701的目标是区分真实的质量属性值50和模拟的质量属性值50，数学上可表示为：
Figure PCTCN2019085055-appb-000002
其中，p(x)is真实的质量属性值50的分布。上式中第一部分将针对真实的质量属性值50的输出最小化；第二部分将针对生成器702输出的模拟的质量属性值50的输出最大化；第三部分为梯度惩罚损失(可参见2017年Gulrajani，Ishaan，Faruk Ahmed，Martin Arjovsky，Vincent Dumoulin，and Aaron C.Courville在《神经信息处理系统进展》(Advances in Neural Information Processing Systems)上第5767至5777页发表的文章《改进的Wasserstein GANs 训练》(Improved training of wasserstein gans)，其中ω为一个超参数(hyperparameter)，通常取值为10。p(x’)沿着真实的质量属性值50和模拟的质量属性值50之间的直线均匀取值，以避免训练过程中梯度消失或者激增。
可以通过多次迭代交替地最小化生成器702和判别器701的目标函数来训练CGAN模型70，直至判别器701无法分区分真实的质量属性值50和模拟的质量属性值50。然后可以使用生成器702生成大量在任意的加工参数40和任意的周围环境条件数据90下的模拟的质量属性值50。据此，模拟的质量属性值50的分布与真实的质量属性值50的分布接近。图7示出了在相同的CNC加工参数和周围环境条件下，从蜗轮制造厂采集的实际工件的质量数据值50和由CGAN模型生成的模拟的质量属性值50的分布。图中把高维的质量属性值通过降维方法映射到二维方便显示。其中，实心的圆代表真实的质量属性值50，而虚心的圆圈代表模拟的质量属性值50。
S203：确定第二关系。
在步骤S203中，可根据复数个工件20中每一个工件20的质量属性值以及质量检验结果数据，确定生产设备10所加工工件的质量检验结果数据60与质量属性值50之间的第二关系。
若质量检验结果数据60为通过质量检验或未通过质量检验，可使用分类器80来确定第二关系。实际上，只要能够达到最佳的分类精度，就可以选择任何分类器，如随机森林、支持向量机，甚至可以在本步骤中使用预定义的可接受范围来检查质量属性值50。具体来说，分类器80的输入是步骤S201中获取的工件20的质量属性值50，输出是工件20的质量检验结果数据60。分类器80根据步骤S201中获取的真实的数据进行训练。训练结束后，可利用分类器80对模拟的质量属性值50是否通过质量检验进行分类，这可以看作是一个虚拟的质量测试。
图5为本发明实施例提供的验证方法的流程图。该验证方法500可由图1中的验证装置302来执行。如图5所示，可包括如下步骤：
S501：生成工件数据。
在步骤S501中，可根据生产设备10所加工的复数个工件20中每一个工件20的加工条件数据40、质量属性值50，以及第一关系生成复数个第一模拟工件20的质量属性值50。
可选地，若通过训练CGAN模型70得到第一关系，则这里可利用CGAN模型70中的生成器702生成大量个模拟的质量属性值50。
S502：将步骤S501中生成的第一模拟工件20与获取的生产设备10所加工的复数个工件 20的质量属性值的分布相比较，以确定第一关系的准确性。
可选地，可使用Kolmogorov-Smirnov检验定量测量真实的质量属性值50和模拟的质量属性值50之间的相似性，以确保模拟的质量属性值50能够充分代表真实的质量属性值50。在统计学中，Kolmogorov-Smirnov检验(k-s检验或k s检验)是一种非参数检验，可用于比较两个样本的连续一维概率分布是否相等。这里，使用测试的多变量泛化来检查模拟的质量属性值50和真实的质量属性值50之间的差异是否具有统计学意义(参见Jerome H.Friedman和Lawrence C.Rafsky在1979年于《统计记录》(The annals of Statistics)上第679至717页发表的文章《Wald-Wolfowitz和Smirnov两个样本检验的多变量泛化》(Multivariate generalizations of the Wald-Wolfowitz and Smirnov two-sample tests))。在确定了模拟的质量属性值50与真实的质量属性值50非常接近后，可以进一步利用模拟的质量属性值50进行质量分析。
在得到了上述第一关系和第二关系之后，即可在任意周围环境条件下，在任意设定的加工参数下，评估工件的质量检验通过率。其中，一种寻求生产设备的加工参数优化设置的暴力方法是枚举所有可能的参数设置，并选择质量检验通过率最高的参数。然而，这种方法成本太高，有时是不可接受的，尤其是当周围环境条件变化迅速，且需要动态调整最佳加工参数的情况下。因此，本发明实施例中还提供了一种加工参数设置方法，如图6所示。
图6为本发明实施例提供的加工参数设置方法的流程图。该加工参数设置方法600可由图1中的加工参数设置装置303来执行。如图6所示，该方法可包括如下步骤：
S601：生成加工参数40的集合。
在步骤S601中，在给定的周围环境条件数据下，生成一个生产设备10的加工参数40的集合，其中，可以随机生成该加工参数40的集合。以CNC为例，随机生成M个加工参数40，其中M为正整数。
S602：生成质量属性值50。
在步骤S602中，对于步骤S601中生成的集合中的每一个元素，根据给定的周围环境条件数据90、该元素以及第一关系，生成复数个第二模拟工件20的质量属性值50。仍以CNC为例，在给定周围环境条件数据90的情况下，针对前述步骤S601中随机生成的M个加工参数40中的每一个，利用CGAN模型70分别生成大量的质量属性值50。
S603：确定质量检验结果数据60。
在步骤S603中，对于集合中的每一个元素，根据第二关系以及生成的复数个第二模拟工 件20的质量属性值50，确定每一个第二模拟工件20的质量检验结果数据60。仍以CNC为例，针对前述步骤S601中随机生成的M个加工参数40中的每一个，利用前述的第二关系，根据步骤S602中生成的该加工参数40所对应的大量的质量属性值50，生成每一个质量属性值50所对应的质量检验结果数据60。
S604：统计质量检验通过率。
在步骤S604中，对于集合中的每一个元素，分别根据确定的该元素所对应的复数个第二模拟工件20的质量检验结果数据60统计该元素所对应的质量检验通过率。仍以CNC为例，对于步骤S601中随机生成的M个加工参数40中的每一个，根据该加工参数40所对应的大量的质量检验结果数据60，统计采用该加工参数40，CNC所加工工件的质量检验通过率。
在步骤S604之后，重复步骤S605至步骤S608，直至满足预设条件。可根据实际情况设置该预设条件。比如：迭代次数达到最大迭代次数；再比如：计算得到的生产设备10所加工工件的质量检验通过率达到通过率阈值；又比如：迭代次数达到最大迭代次数或计算得到的生产设备10所加工工件的质量检验通过率达到通过率阈值这两个条件之一满足即跳出该循环。
S605：在集合中加入一个新的元素。
在步骤S605中，在集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大。其中，可根据集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程(Guassian Process，GP)(参见Rasmussen，Carl Edward于2004年在《机器学习中的高斯过程》第63至71页上发表的《机器学习中的高斯过程》)，并利用高斯过程计算得到新的元素。仍以CNC为例，可拟合一个高斯过程，以集合中的所有元素，即每一个加工参数40作为高斯过程的输入，将每一个加工参数40所对应的质量检验通过率作为高斯过程的输出。
具体地，可使用贝叶斯优化(参见Frazier，Peter I.于2018年在arXiv preprint arXiv：1807.02811上发表的《贝叶斯优化教程》(A Tutorial on Bayesian Optimization))来有效地搜索加工参数40的候选空间。然后在给定的集合中各个加工参数40设置下，使用高斯过程来估计所有未搜索设置的工件质量检验通过率的后验分布：
P(y|x，θ，D)
其中，x是未搜索的设置，θ是高斯过程的参数，D是各搜索设置及各自对应的工件质量检验通过率的估计值。然后，我们可以通过选择一个最大限度地提高质量检验
Figure PCTCN2019085055-appb-000003
通过率期望值的设置来有效搜索加工参数40的空间，计算公式如下：
Figure PCTCN2019085055-appb-000004
其中，
Figure PCTCN2019085055-appb-000005
表示已搜索设置中最高的质量检验通过率。上述公式的详解可参见前述的《贝叶斯优化教程》。
利用上述贝叶斯优化方法，在特定的周围环境条件数据90下，实现了加工参数40的最优化配置。
S606：针对该新的元素生成质量属性值。
在步骤S606中，根据给定的周围环境条件数据、该新的元素以及上述第一关系，确定依据该新的元素所代表的加工参数所生产的复数个第三模拟工件20中每一个的质量属性50。
S607：针对新的元素确定质量检验结果数据。
在步骤S607中，对于该新的元素，针对复数个第三模拟工件20中的每一个，根据上述第二关系，以及确定的该复数个第三模拟工件20的质量属性值，确定该复数个第三模拟工件20中每一个的质量检验结果数据60。
S608：针对该新的元素统计质量检验通过率。
在步骤S608中，对于集合中加入的该新的元素，根据确定的复数个第三模拟工件20中每一个的质量检验结果数据统计质量检验通过率。S609：判断是否满足上述预设条件。如满足上述预设条件，则跳出循环执行步骤S610，否则返回步骤S605。
S610：确定加工参数的最优值。
确定统计得到的所有质量检验通过率最大值所对应的元素，作为在给定的周围环境条件数据下生产设备10的加工参数的最优值。
在图6所示的流程中，迭代次数为N，初始设置为0。N以及步骤S601中生成的加工参数40的集合中元素的个数M的值可需要考虑的加工参数40的数量而定。以CNC为例，CNC加工参数的数量通常小于10，因此，这里可以考虑将M设置为5，N设置为10。具体可根据实验调整M和N的值。
图8为本发明实施例提供的工件数据处理装置的结构示意图。该工件数据处理装置30可包括：一个工件数据处理装置301，可选地，还可包括一个验证装置302和/或一个加工参数设置装置303。
其中，工件数据处理装置301可包括：一个工件数据获取模块301a，被配置为获取由一个生产设备10所加工的复数个工件20中每一个工件20的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件20的加工条件数据包括：生产设备10在加工该工件20时所 使用的加工参数以及生产设备10在加工该工件20时的周围环境条件数据；一个第一关系确定模块301b，被配置为根据复数个工件20中每一个工件20的加工条件数据和质量属性值，确定生产设备10所加工工件的质量属性值与生产设备10在加工工件时的周围环境条件数据和生产设备10的加工参数之间的第一关系；一个第二关系确定模块301c，被配置为根据复数个工件20中每一个工件20的质量属性值以及质量检验结果数据，确定生产设备10所加工工件的质量检验结果数据与质量属性值与之间的第二关系。
可选地，工件数据获取模块301a被具体配置为：获取由一个生产设备10所加工的复数个工件20中每一个工件20的加工条件数据、质量属性值以及质量检验结果数据，以使得获取的所有数据尽量覆盖各种周围环境条件数据和加工参数的组合。
可选地，第一关系确定模块301b，被具体配置为：将复数个工件20中每一个工件20的加工条件数据作为一个CGAN模型中生成器702的一个输入向量；将复数个工件20中的每一个工件20的加工条件数据作为CGAN模型的判别器701的一个输入向量；将复数个工件20中的每一个工件20的质量属性值或CGAN模型的生成器的输出向量702a，作为CGAN模型中判别器701的另一个输入向量；训练CGAN模型，将训练得到的CGAN模型的生成器701作为第一关系。
其中，验证装置302可包括：一个模拟工件数据生成模块302a，被配置为根据生产设备10所加工的复数个工件20中每一个工件20的加工条件数据、质量属性值，以及第一关系生成复数个第一模拟工件20的质量属性值；一个数据比较模块302b，被配置为将复数个第一模拟工件20与获取的生产设备10所加工的复数个工件20的质量属性值的分布相比较，以确定第一关系的准确性。
其中，加工参数设置装置303可包括：
一个加工参数集合建立模块303a，被配置为在给定的周围环境条件数据下，生成一个生产设备10的加工参数的集合；
一个调整模块303b，被配置为对于集合中的每一个元素，根据给定的周围环境条件数据、该元素以及第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件20中每一个的质量属性值，其中，第一关系为生产设备10所加工工件的质量属性值与生产设备10在加工工件时的周围环境条件数据和生产设备10的加工参数之间的关系；对于集合中的每一个元素，针对复数个第二模拟工件20中的每一个，根据第二关系以及确定的该第二模拟工件20的质量属性值，确定该第二模拟工件20的质量检验结果数据，其中，第二关系为生产设备10所加工工件的质量检验结果数据与质量属性值之间的关系；对于集合中的每一个元素， 根据确定的复数个第二模拟工件20中每一个的质量检验结果数据统计质量检验通过率；
调整模块303b，还被配置为重复下述过程直至满足预设条件，其中预设条件包括：迭代次数达到最大迭代次数或计算得到的生产设备10所加工工件的质量检验通过率达到通过率阈值：在集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；对于集合中加入的该新的元素，根据给定的周围环境条件数据、该新的元素以及第一关系，确定依据该新的元素所代表的加工参数所生产的复数个第三模拟工件20中每一个的质量属性值；对于集合中加入的该新的元素，针对复数个第三模拟工件20中的每一个，根据第二关系以及确定的复数个第三模拟工件20的质量属性值，确定复数个第三模拟工件20中每一个的质量检验结果数据；对于集合中加入的该新的元素，根据确定的复数个第三模拟工件20中每一个的质量检验结果数据统计质量检验通过率；
一个最优加工参数确定模块303c，被配置为确定统计得到的所有质量检验通过率最大值所对应的元素，作为在给定的周围环境条件数据下生产设备10的加工参数的最优值。
可选地，调整模块303b在集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大时，具体被配置为：根据集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程；利用高斯过程计算得到新的元素。
图9为本发明实施例提供的工件数据处理装置30的又一结构示意图。如图9所示，在此结构下，工件数据处理装置30可包括至少一个存储器304，用于存放计算机可读代码；至少一个处理器305，被配置为执行至少一个存储器304存储的计算机可读代码，执行前述的方法200，可选地，还可执行前述的方法500和/或方法600。其中，图8中示出的各个模块可视为存储器304中存储的计算机可读代码编写的程序模块，当这些程序模块被处理器305调用时，能够执行前述方法200、500和/或600。工件数据处理装置30还可包括一个I/O接口306，该I/O接口可连接鼠标、显示器等外部设备。可选地，至少一个存储器304，至少一个处理器305以及I/O接口306之间可通过总线307通信。
此外，本发明实施例还提供一种计算机可读介质，该计算机可读介质存储有计算机可读代码，当该计算机可读代码被至少一个处理器执行时，实现前述方法200、500和/或600。
综上，本发明实施例提供一种工件数据处理方法、装置和计算机可读介质，用于准确确定生产设备的加工参数、周围环境条件数据与工件的质量检验结果之间的关系。进而能够依据该关系确定最优的加工参数。
本发明实施例提供的工件数据处理方法是纯数据驱动的，与由人的经验驱动的方法相比，显著提高了为工件加工设置最佳加工参数的过程的准确性、效率和通用性。
本发明实施例所提供的方案具有成本效益，预测在给定的加工参数和周围环境条件下生产工件的通过率而构建数学模型所需的数据量。采用本发明实施例提供的方案仅需要搜索加工参数候选空间的一小部分，以找到最佳的加工参数。
本发明实施例中，利用CGAN模型来模拟在给定的加工参数下工件的质量属性值，然后利用分类器来执行虚拟的质量检验以确定工件的质量检验通过率。采用本发明实施例所提供的方案可仅适用少量的工件数据来确定生产设备的加工参数与质量检验结果数据之间的关系。与仅依靠人的经验的方法相比，确定的关系更准确。
本发明实施例中，可依靠贝叶斯优化方法来自动有效地找到生产设备的加工参数的最佳设置，从而在任意周围环境条件下获得最高的工件质量检验通过率。
需要说明的是，上述各流程和各系统结构图中不是所有的步骤和模块都是必须的，可以根据实际的需要忽略某些步骤或模块。各步骤的执行顺序不是固定的，可以根据需要进行调整。上述各实施例中描述的系统结构可以是物理结构，也可以是逻辑结构，即，有些模块可能由同一物理实体实现，或者，有些模块可能分由至少两个物理实体实现，或者，可以由至少两个独立设备中的某些部件共同实现。
以上各实施例中，硬件单元可以通过机械方式或电气方式实现。例如，一个硬件单元可以包括永久性专用的电路或逻辑(如专门的处理器，现场可编程门阵列(Field-Programmable Gate Array，FPGA)或专用集成电路(Application Specific Integrated Circuits，ASIC)等)来完成相应操作。硬件单元还可以包括可编程逻辑或电路(如通用处理器或其它可编程处理器)，可以由软件进行临时的设置以完成相应操作。具体的实现方式(机械方式、或专用的永久性电路、或者临时设置的电路)可以基于成本和时间上的考虑来确定。
上文通过附图和优选实施例对本发明实施例进行了详细展示和说明，然而本发明实施例不限于这些已揭示的实施例，基于上述实施例本领域技术人员可以知晓，可以组合上述不同 实施例中的代码审核手段得到本发明更多的实施例，这些实施例也在本发明实施例的保护范围之内。

Claims (18)

  1. 工件数据处理方法(200)，其特征在于，包括：
    获取由一个生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件(20)的加工条件数据包括：所述生产设备(10)在加工该工件(20)时所使用的加工参数以及所述生产设备(10)在加工该工件(20)时的周围环境条件数据；
    根据所述复数个工件(20)中每一个的加工条件数据和质量属性值，确定所述生产设备(10)所加工工件的质量属性值与所述生产设备(10)在加工工件时的周围环境条件数据和所述生产设备(10)的加工参数之间的第一关系；
    根据所述复数个工件(20)中每一个的质量属性值以及质量检验结果数据，确定所述生产设备(10)所加工工件的质量检验结果数据与质量属性值之间的第二关系。
  2. 如权利要求1所述的方法(200)，其特征在于，获取由一个生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值以及质量检验结果数据，包括：
    获取由所述生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值以及质量检验结果数据，以使得获取的所有数据尽量覆盖各种周围环境条件数据和加工参数的组合。
  3. 如1或2所述的方法(200)，其特征在于，确定所述第一关系，包括：
    将所述复数个工件(20)中每一个的加工条件数据作为一个CGAN模型中生成器(702)的一个输入向量；
    将所述复数个工件(20)中的每一个的加工条件数据作为所述CGAN模型的判别器(701)的一个输入向量；
    将所述复数个工件(20)中的每一个工件的质量属性值或所述CGAN模型的生成器的输出向量(702a)，作为所述CGAN模型中判别器(701)的另一个输入向量；
    训练所述CGAN模型，将训练得到的所述CGAN模型的生成器(702)作为所述第一关系。
  4. 如权利要求1～3任一项所述的方法(200)，其特征在于，还包括：
    根据所述生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值，以及所述第一关系生成复数个第一模拟工件(20)中每一个的质量属性值；
    将所述复数个第一模拟工件(20)与获取的所述生产设备(10)所加工的所述复数个工件(20)的质量属性值的分布相比较，以确定所述第一关系的准确性。
  5. 如权利要求1～4任一项所述的方法(200)，其特征在于，还包括：
    在给定的周围环境条件数据下，生成一个所述生产设备(10)的加工参数的集合；
    对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及所述第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件(20)中每一个的质量属性值；
    对于所述集合中的每一个元素，针对复数个第二模拟工件(20)中的每一个，根据所述第二关系，以及确定的该第二模拟工件(20)的质量属性值，确定该第二模拟工件(20)的质量检验结果数据；
    对于所述集合中的每一个元素，根据确定的复数个第二模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备(10)所加工工件的质量检验通过率达到通过率阈值，
    在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；
    对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所加工的复数个第三模拟工件(20)中每一个的质量属性值；
    对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件(20)中的每一个，根据所述第二关系，以及确定的所述复数个第三模拟工件(20)的质量属性值，确定所述复数个第三模拟工件(20)中每一个的质量检验结果数据；
    对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备(10)的加工参数的最优值。
  6. 如权利要求5所述的方法(200)，其特征在于，在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的 最大值相比增幅最大，包括：
    根据所述集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程；
    利用所述高斯过程计算得到所述新的元素。
  7. 工件数据处理装置(30)，其特征在于，包括：
    一个工件数据获取模块(301a)，被配置为获取由一个生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值以及质量检验结果数据，其中一个工件(20)的加工条件数据包括：所述生产设备(10)在加工该工件(20)时所使用的加工参数以及所述生产设备(10)在加工该工件(20)时的周围环境条件数据；
    一个第一关系确定模块(301b)，被配置为根据所述复数个工件(20)中每一个的加工条件数据和质量属性值，确定所述生产设备(10)所加工工件的质量属性值与所述生产设备(10)在加工工件时的周围环境条件数据和所述生产设备(10)的加工参数之间的第一关系；
    一个第二关系确定模块(301c)，被配置为根据所述复数个工件(20)中每一个的质量属性值以及质量检验结果数据，确定所述生产设备(10)所加工工件的质量检验结果数据与质量属性值之间的第二关系。
  8. 如权利要求7所述的装置(30)，其特征在于，所述工件数据获取模块(301a)被具体配置为：获取由所述生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值以及质量检验结果数据，以使得获取的所有数据尽量覆盖各种周围环境条件数据和加工参数的组合。
  9. 如7或8所述的装置(30)，其特征在于，所述第一关系确定模块(301b)，被具体配置为：
    将所述复数个工件(20)中每一个的加工条件数据作为一个CGAN模型中生成器(702)的一个输入向量；
    将所述复数个工件(20)中的每一个的加工条件数据作为所述CGAN模型的判别器(701)的一个输入向量；
    将所述复数个工件(20)中的每一个的质量属性值或所述CGAN模型的生成器的输出向量(702a)，作为所述CGAN模型中判别器(701)的另一个输入向量；
    训练所述CGAN模型，将训练得到的所述CGAN模型的生成器(702)作为所述第一关系。
  10. 如权利要求7～9任一项所述的装置(30)，其特征在于，还包括：
    一个模拟工件数据生成模块(302a)，被配置为根据所述生产设备(10)所加工的复数个工件(20)中每一个的加工条件数据、质量属性值，以及所述第一关系生成复数个第一模拟工件(20)中每一个的质量属性值；
    一个数据比较模块(302b)，被配置为将所述复数个第一模拟工件(20)与获取的所述生产设备(10)所加工的所述复数个工件(20)的质量属性值的分布相比较，以确定所述第一关系的准确性。
  11. 如权利要求7～10任一项所述的装置(30)，其特征在于，还包括：
    一个加工参数集合建立模块(303a)，被配置为在给定的周围环境条件数据下，随机生成一个所述生产设备(10)的加工参数的集合；
    一个调整模块(303b)，被配置为：
    对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及所述第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件(20)中每一个的质量属性值；
    对于所述集合中的每一个元素，针对复数个第二模拟工件(20)中的每一个，根据所述第二关系，以及确定的该第二模拟工件(20)的质量属性值，确定该第二模拟工件(20)的质量检验结果数据；
    对于所述集合中的每一个元素，根据确定的复数个第二模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备(10)所加工工件的质量检验通过率达到通过率阈值，
    在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；
    对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所加工的复数个第三模拟工件(20)中每一个的质量属性值；
    对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件(20)中的每一个，根据所述第二关系，以及确定的所述复数个第三模拟工件(20)的质量属性值，确定所述复数个第三模拟工件(20)中每一个的质量检验结果数据；
    对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    一个最优加工参数确定模块(303c)，被配置为确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备(10)的加工参数的最优值。
  12. 如权利要求11所述的装置(30)，其特征在于，所述调整模块(303b)在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大时，具体被配置为：
    根据所述集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程；
    利用所述高斯过程计算得到所述新的元素。
  13. 工件数据处理装置(30)，其特征在于，包括：
    至少一个存储器(304)，用于存放计算机可读代码；
    至少一个处理器(305)，用于执行所述至少一个存储器(304)存储的所述计算机可读代码，执行如权利要求1～6任一项所述的方法。
  14. 加工参数设置方法(600)，其特征在于，包括：
    在给定的周围环境条件数据下，生成一个所述生产设备(10)的加工参数的集合；
    对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及所述第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件(20)中每一个的质量属性值，其中，所述第一关系为所述生产设备(10)所加工工件的质量属性值与所述生产设备(10)在加工工件时的周围环境条件数据和所述生产设备(10)的加工参数之间的关系；
    对于所述集合中的每一个元素，针对复数个第二模拟工件(20)中的每一个，根据所述第二关系，以及确定的该第二模拟工件(20)的质量属性值，确定该第二模拟工件(20)的 质量检验结果数据，其中，所述第二关系为所述生产设备(10)所加工工件的质量检验结果数据与质量属性值之间的关系；
    对于所述集合中的每一个元素，根据确定的复数个第二模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备(10)所加工工件的质量检验通过率达到通过率阈值，
    在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；
    对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所加工的复数个第三模拟工件(20)中每一个的质量属性值；
    对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件(20)中的每一个，根据所述第二关系，以及确定的所述复数个第三模拟工件(20)的质量属性值，确定所述复数个第三模拟工件(20)中每一个的质量检验结果数据；
    对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备(10)的加工参数的最优值。
  15. 如权利要求14所述的方法(600)，其特征在于，在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大，包括：
    根据所述集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程；
    利用所述高斯过程计算得到所述新的元素。
  16. 加工参数设置装置(303)，其特征在于，包括：
    一个加工参数集合建立模块(303a)，被配置为在给定的周围环境条件数据下，生成一个所述生产设备(10)的加工参数的集合；
    一个调整模块(303b)，被配置为：
    对于所述集合中的每一个元素，根据所述给定的周围环境条件数据、该元素以及第一关系，确定依据该元素代表的加工参数所加工的复数个第二模拟工件(20)中每一个的质量属性值，其中，所述第一关系为所述生产设备(10)所加工工件的质量属性值与所述生产设备(10)在加工工件时的周围环境条件数据和所述生产设备(10)的加工参数之间的关系；
    对于所述集合中的每一个元素，针对复数个第二模拟工件(20)中的每一个，根据第二关系以及确定的该第二模拟工件(20)的质量属性值，确定该第二模拟工件(20)的质量检验结果数据，其中，所述第二关系为所述生产设备(10)所加工工件的质量检验结果数据与质量属性值之间的关系；
    对于所述集合中的每一个元素，根据确定的复数个第二模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    重复下述过程直至满足预设条件，其中所述预设条件包括：迭代次数达到最大迭代次数或计算得到的所述生产设备(10)所加工工件的质量检验通过率达到通过率阈值，
    在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大；
    对于所述集合中加入的该新的元素，根据所述给定的周围环境条件数据、该新的元素以及所述第一关系，确定依据该新的元素所代表的加工参数所生产的复数个第三模拟工件(20)中每一个的质量属性值；
    对于所述集合中加入的该新的元素，针对所述复数个第三模拟工件(20)中的每一个，根据所述第二关系以及确定的所述复数个第三模拟工件(20)的质量属性值，确定所述复数个第三模拟工件(20)中每一个的质量检验结果数据；
    对于所述集合中加入的该新的元素，根据确定的所述复数个第三模拟工件(20)中每一个的质量检验结果数据统计质量检验通过率；
    一个最优加工参数确定模块(303c)，确定统计得到的所有质量检验通过率最大值所对应的元素，作为在所述给定的周围环境条件数据下所述生产设备(10)的加工参数的最优值。
  17. 如权利要求16所述的装置(303)，其特征在于，所述调整模块(303b)在所述集合中加入一个新的元素，以使依据新的元素统计得到的质量检验通过率的期望值与以往统计得到的质量检验通过率的最大值相比增幅最大时，具体被配置为：
    根据所述集合中每一个元素和以往统计得到的该元素所对应的质量检验通过率拟合一个高斯过程；
    利用所述高斯过程计算得到所述新的元素。
  18. 计算机可读介质，其特征在于，所述计算机可读介质存储有计算机可读代码，当所述计算机可读代码被至少一个处理器执行时，执行如权利要求1～6，任一项或14、15所述的方法。
PCT/CN2019/085055 2019-04-29 2019-04-29 生产设备的加工参数设置方法、装置和计算机可读介质 WO2020220221A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
US17/606,774 US11982996B2 (en) 2019-04-29 2019-04-29 Method and apparatus for configuring processing parameters of production equipment, and computer-readable medium
CN201980091385.8A CN113646714A (zh) 2019-04-29 2019-04-29 生产设备的加工参数设置方法、装置和计算机可读介质
PCT/CN2019/085055 WO2020220221A1 (zh) 2019-04-29 2019-04-29 生产设备的加工参数设置方法、装置和计算机可读介质
EP19927216.2A EP3951523A4 (en) 2019-04-29 2019-04-29 METHOD AND DEVICE FOR CONFIGURING PROCESSING PARAMETERS OF A PRODUCTION PLANT AND COMPUTER READABLE MEDIUM
EP23200043.0A EP4276554A3 (en) 2019-04-29 2019-04-29 Method and apparatus for configuring processing parameters of production equipment, and computer-readable medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/085055 WO2020220221A1 (zh) 2019-04-29 2019-04-29 生产设备的加工参数设置方法、装置和计算机可读介质

Publications (1)

Publication Number Publication Date
WO2020220221A1 true WO2020220221A1 (zh) 2020-11-05

Family

ID=73028696

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/085055 WO2020220221A1 (zh) 2019-04-29 2019-04-29 生产设备的加工参数设置方法、装置和计算机可读介质

Country Status (4)

Country Link
US (1) US11982996B2 (zh)
EP (2) EP4276554A3 (zh)
CN (1) CN113646714A (zh)
WO (1) WO2020220221A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836826A (zh) * 2021-11-25 2021-12-24 深圳市裕展精密科技有限公司 关键参数确定方法、装置、电子装置及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11901204B2 (en) * 2020-05-22 2024-02-13 Applied Materials, Inc. Predictive wafer scheduling for multi-chamber semiconductor equipment
CN117311295B (zh) * 2023-11-28 2024-01-30 深圳百通玄武技术有限公司 基于无线网络设备的生产质量提升方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101488024A (zh) * 2009-01-23 2009-07-22 秦皇岛烟草机械有限责任公司 烟草加工过程参数的在线质量评估与实时智能控制方法
CN103576646A (zh) * 2013-10-07 2014-02-12 红塔烟草(集团)有限责任公司 一种提高卷烟制丝加工质量数据分析系统适用性的动态配置方法
CN107122609A (zh) * 2017-04-28 2017-09-01 电子科技大学 一种基于质量特性基因理论的机电产品质量评价方法
CN107194826A (zh) * 2017-06-16 2017-09-22 北京航空航天大学 一种基于质量状态任务网的制造系统健康诊断与预测方法

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR920002268A (ko) * 1990-07-17 1992-02-28 유끼노리 가까즈 인텔리젠트가공장치
DE10207048B4 (de) * 2002-02-20 2006-01-19 Hans Adler Ohg Verfahren zum Herstellen von rohen Schinken
WO2007067645A2 (en) * 2005-12-05 2007-06-14 Fisher-Rosemount Systems, Inc. Multi-objective predictive process optimization with concurrent process simulation
JP5418408B2 (ja) * 2010-05-31 2014-02-19 富士通株式会社 シミュレーションパラメータ校正方法、装置及びプログラム
DE102011077568B4 (de) * 2011-06-15 2023-12-07 Dmg Mori Ultrasonic Lasertec Gmbh Werkzeugmaschine, Werkstückbearbeitungsverfahren
US9047423B2 (en) * 2012-01-12 2015-06-02 International Business Machines Corporation Monte-Carlo planning using contextual information
DE102012213504A1 (de) * 2012-07-31 2014-02-06 Robert Bosch Gmbh Verfahren zum Fertigen von Werkstücken, bei dem ein Bearbeitungsparameter eines Bearbeitungsschrittes abhängig von einer bereits geänderten Werkstückeigenschaft festgelegt wird
TWI539298B (zh) * 2015-05-27 2016-06-21 國立成功大學 具取樣率決定機制的量測抽樣方法 與其電腦程式產品
JP6453805B2 (ja) 2016-04-25 2019-01-16 ファナック株式会社 製品の異常に関連する変数の判定値を設定する生産システム
TWI625615B (zh) * 2016-11-29 2018-06-01 財團法人工業技術研究院 預測模型建立方法及其相關預測方法與電腦程式產品
CN108700852B (zh) * 2017-01-27 2021-07-16 三菱动力株式会社 模型参数值推定装置及推定方法、记录介质、模型参数值推定系统
US10678216B2 (en) * 2017-02-28 2020-06-09 Sap Se Manufacturing process data collection and analytics
CN107292422B (zh) * 2017-05-27 2020-07-28 西安交通大学 一种信息-物理-社交融合空间下实现智能制造的S2ensor
US11676247B2 (en) * 2017-07-31 2023-06-13 Institut Pasteur Method, device, and computer program for improving the reconstruction of dense super-resolution images from diffraction-limited images acquired by single molecule localization microscopy
JP6693919B2 (ja) * 2017-08-07 2020-05-13 ファナック株式会社 制御装置及び機械学習装置
CN109242105B (zh) * 2018-08-17 2024-03-15 第四范式（北京）技术有限公司 代码优化方法、装置、设备及介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101488024A (zh) * 2009-01-23 2009-07-22 秦皇岛烟草机械有限责任公司 烟草加工过程参数的在线质量评估与实时智能控制方法
CN103576646A (zh) * 2013-10-07 2014-02-12 红塔烟草(集团)有限责任公司 一种提高卷烟制丝加工质量数据分析系统适用性的动态配置方法
CN107122609A (zh) * 2017-04-28 2017-09-01 电子科技大学 一种基于质量特性基因理论的机电产品质量评价方法
CN107194826A (zh) * 2017-06-16 2017-09-22 北京航空航天大学 一种基于质量状态任务网的制造系统健康诊断与预测方法

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
FRAZIER, PETER I.: "A Tutorial on Bayesian Optimization", ARXIV: 1807.02811, 2018
GULRAJANI, ISHAANFARUK AHMEDMARTIN ARJOVSKYVINCENT DUMOULINAARON C. COURVILLE, ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS, 2017
JEROME H. FRIEDMANLAWRENCE C. RAFSKY: "Multivariate generalizations of the Wald-Wolfowitz and Smirnov two-sample tests", THE ANNALS OF STATISTICS, 1979, pages 679 - 717
RASMUSSEN, CARL EDWARD: "Gaussian Processes in Machine Learning", GAUSSIAN PROCESSES IN MACHINE LEARNING, 2004, pages 63 - 71, XP055744939, DOI: 10.1007/978-3-540-28650-9_4

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836826A (zh) * 2021-11-25 2021-12-24 深圳市裕展精密科技有限公司 关键参数确定方法、装置、电子装置及存储介质

Also Published As

Publication number Publication date
EP3951523A1 (en) 2022-02-09
CN113646714A (zh) 2021-11-12
EP3951523A4 (en) 2023-06-21
EP4276554A2 (en) 2023-11-15
US11982996B2 (en) 2024-05-14
EP4276554A3 (en) 2024-05-22
US20220206469A1 (en) 2022-06-30

Similar Documents

Publication Publication Date Title
Sun et al. Fast implementation of DeLong’s algorithm for comparing the areas under correlated receiver operating characteristic curves
WO2020220221A1 (zh) 生产设备的加工参数设置方法、装置和计算机可读介质
Ma et al. Hand joints-based gesture recognition for noisy dataset using nested interval unscented Kalman filter with LSTM network
Arunadevi et al. A study of classification algorithms using Rapidminer
JP2008538041A (ja) 局所的近傍点ラプラス固有マップに基づくデータ分類の一部学習済学習機械
Yang Monitoring and diagnosing of mean shifts in multivariate manufacturing processes using two-level selective ensemble of learning vector quantization neural networks
Arif et al. Cascade quality prediction method using multiple PCA+ ID3 for multi-stage manufacturing system
JP2023145545A (ja) 製造プロセスのためのシステム、方法、および媒体
JP6283112B2 (ja) データに基づく関数モデルを定めるための方法及び装置
Kozłowski et al. Integrating advanced measurement and signal processing for reliability decision-making
US20210311440A1 (en) Systems, Methods, and Media for Manufacturing Processes
CN116738920B (zh) 一种三维几何内核的芯片3d设计方法及系统
Sinha et al. Building a scalable and interpretable Bayesian deep learning framework for quality control of free form surfaces
US20210192779A1 (en) Systems, Methods, and Media for Manufacturing Processes
CN113590458A (zh) 用于检查技术系统的方法和设备
WO2012092132A2 (en) Determining the uniqueness of a model for machine vision
JP2023516776A (ja) 製造プロセスのためのシステム、方法及び媒体
CN108021876B (zh) 一种基于参数辨识的压气机故障诊断方法
Shoukat et al. Approximate GP inference for nonlinear dynamical system identification using data-driven basis set
Corchado et al. Optimizing the operating conditions in a high precision industrial process using soft computing techniques
Stollfuss et al. MLP-Supported Mathematical Optimization of Simulation Models: Investigation into the Approximation of Black Box Functions of Any Simulation Model with MLPs with the Aim of Functional Analysis.
Elvikis et al. Scheduling and location (ScheLoc): makespan problem with variable release dates
US20220383103A1 (en) Hardware accelerator method and device
EP4168941A1 (en) Systems, methods, and media for manufacturing processes
Nugroho et al. Analysis of Optimization Techniques in 6D Pose Estimation Approaches using RGB Images on Multiple Objects with Occlusion

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19927216

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019927216

Country of ref document: EP

Effective date: 20211026