JP2021010980A - Grinding state monitoring method, grinding state monitoring program and apparatus - Google Patents

Abstract

To improve accuracy in prediction of grinding quality.SOLUTION: A grinding quality prediction device acquires information indicating transition of a power value that drives a grinding wheel and quality of a grinding object when the grinding object is ground using the grinding wheel. The grinding quality prediction device extracts a power value of a common part of a plurality of grinding objects to be ground using the grinding wheel that is a power value when the grinding wheel grinds the common part, from the transition of the power value. The grinding quality prediction device generates a quality prediction model for predicting a grinding state by using data based on the power value of the common part and the information indicating the quality of the grinding object.SELECTED DRAWING: Figure 2

Description

The present invention relates to a grinding condition monitoring method, a grinding condition monitoring program and an apparatus.

In the manufacturing industry of automobiles and the like, the quality is kept constant by rotating a grinding wheel at a high speed and performing a grinding process for grinding a workpiece indicating an object to be ground such as a camshaft. Repeated grinding reduces the crushing and balance of the grinding wheel, which leads to deterioration of the quality of the ground work. Therefore, in order to keep the quality constant, the grinding wheel is dressed regularly.

Generally, the dress timing of the grinding wheel is set to a certain number of times of grinding by the experience of the worker and the trial and error of the worker. In recent years, there has been known a technique of performing dressing by determining that the state of the grinding wheel has deteriorated, such as a decrease in sharpness, when the load current value of the rotary motor for rotating the grinding wheel exceeds a threshold value.

Japanese Unexamined Patent Publication No. 2000-263437 JP-A-2009-006456 Japanese Unexamined Patent Publication No. 51-050083

However, even if the grinding surface of each work is in the same state, it is difficult to accurately predict the surface roughness of the grinding wheel because the load current value at the time of grinding to rotate the grinding wheel varies.

If the determination of the dress timing becomes inaccurate, a state may occur in which the grinding wheel whose quality has not deteriorated is dressed or the grinding process is continuously performed with the grinding wheel whose quality has deteriorated. As a result, the cost of dressing increases due to unnecessary dressing, the life of the grinding wheel is shortened, and the quality of the ground work is deteriorated.

In one aspect, it is an object of the present invention to provide a grinding condition monitoring method, a grinding condition monitoring program and an apparatus capable of improving the prediction accuracy of grinding quality.

In the first proposal, the grinding condition monitoring method is information indicating the transition of the power value that drives the grinding wheel and the quality of the grinding object when the computer grinds the grinding object using the grinding wheel. Execute the process to get and. The grinding state monitoring method is a process in which a computer extracts a power value of a common part, which is a power value when a common part of a plurality of grinding objects to be ground by the grinding wheel is ground, from a transition of the power value. To execute. In the grinding state monitoring method, a computer executes a process of generating a quality prediction model for predicting a grinding state by using data based on a power value of the common portion and information indicating the quality of the grinding object.

According to one embodiment, the prediction accuracy of grinding quality can be improved.

FIG. 1 is a diagram illustrating an overall configuration of a grinding apparatus according to a first embodiment. FIG. 2 is a diagram illustrating a grinding quality prediction device according to the first embodiment. FIG. 3 is a functional block diagram showing a functional configuration of the grinding quality prediction device according to the first embodiment. FIG. 4 is a diagram illustrating an example of grinding data. FIG. 5 is a diagram for explaining transition data of the power value. FIG. 6 is a diagram illustrating an example of learning data. FIG. 7 is a diagram for explaining the specification of the timing for grinding the base circle. FIG. 8 is a diagram illustrating determination of each process. FIG. 9 is a diagram illustrating an example of generating data for learning using the power value at the time of grinding the base circle. FIG. 10 is a diagram illustrating an example of generating data for learning using the power values of the fine research process and the fine research process. FIG. 11 is a diagram illustrating an example of generating data for learning excluding the power value during the Kuken process. FIG. 12 is a diagram illustrating learning of the quality prediction model. FIG. 13 is a flowchart showing the flow of the learning process. FIG. 14 is a flowchart showing the flow of the prediction process. FIG. 15 is a diagram illustrating an example in which the diameter of the grinding wheel is used for learning data. FIG. 16 is a diagram illustrating an example in which the diameter and rotation speed of the camshaft are used for learning data. FIG. 17 is a diagram illustrating a hardware configuration example.

Hereinafter, examples of the grinding condition monitoring method, the grinding condition monitoring program, and the apparatus disclosed in the present application will be described in detail with reference to the drawings. The present invention is not limited to this embodiment. In addition, each embodiment can be appropriately combined within a consistent range.

[overall structure]
FIG. 1 is a diagram illustrating an overall configuration of a grinding apparatus according to a first embodiment. As shown in FIG. 1, in the grinding device, the grinding machine 1 and the grinding quality prediction device 10 are connected to each other via a network N so as to be communicable with each other. The network N can adopt various communication networks such as the Internet and a dedicated line regardless of whether it is wired or wireless.

The grinding machine 1 is a device that grinds a workpiece to be ground and produces a workpiece of desired quality. Specifically, the grinding machine 1 drives the grinding wheel 2 to grind the camshaft 3 which is an example of the work. Further, the grinding machine 1 guarantees the quality by regularly dressing the grinding wheel 2 using the dressing device 4.

The grinding quality prediction device 10 is an example of a computer device that acquires information indicating the grinding process of the grinding machine 1 and the quality of a grinding object and predicts the grinding quality. Specifically, the grinding quality prediction device 10 includes a grindstone shaft power value (hereinafter, may be simply referred to as a power value) indicating a power value for driving or rotating the grinding wheel 2, and a camshaft 3 after grinding. The quality prediction model is trained by inputting the relationship with the quality of. After that, the grinding quality prediction device 10 predicts the state of the grinding wheel 2 using the quality prediction model after learning, and outputs an index of dressing timing to the user or the like.

Generally, when grinding is performed, the specification information of the object to be ground is programmed in the grinding machine 1. Then, before processing, the grinding machine confirms the type of the object to be ground and performs processing according to the specifications according to each. For example, when grinding the camshaft 3, the model information written on the camshaft 3 is read in advance, and the grinding process is performed so as to have the read specifications. A cross section of the camshaft 3 is shown in FIG. During the grinding process, the camshaft 3 is clamped and fixed to the processing machine to perform a rotational operation. The grinding wheel 2 approaches the rotating camshaft 3. The approach follows the grinding process, approaches the camshaft 3 while keeping the depth of cut constant, and performs grinding. The grinding wheel 2 grinds while controlling the coordinates while matching the shapes of the camshaft base circle (the thick line portion of the camshaft 3 in FIG. 1), the top, and the lift portion.

As described above, the grinding machine 1 does not continuously grind only one type of camshaft 3, but also grinds camshafts 3 having different grinding conditions such as a model. Further, since the power value of the grinding wheel 2 differs depending on each model, it is conceivable to learn the quality prediction model for each model. However, since the number of models is enormous, it is not realistic to learn a quality prediction model corresponding to all grinding conditions.

Therefore, the grinding quality prediction device 10 of the first embodiment pays attention to the grinding of the base circle portion of the camshaft 3 which often has the same shape regardless of the model. That is, the grinding quality prediction device 10 generates a model-independent quality prediction model by using the power value when the base circle portion is ground as learning data from the information indicating the grinding quality of each camshaft 3. To do.

[Explanation of Grinding Quality Predictor 10]
FIG. 2 is a diagram for explaining the grinding quality prediction device 10 according to the first embodiment. As shown in FIG. 2, the grinding quality prediction device 10 executes a learning phase and a prediction phase. In the first embodiment, the example in which the grinding quality prediction device 10 executes both phases will be described, but the present invention is not limited to this, and each phase can be executed by a separate device.

As shown in FIG. 2, in the learning phase, the grinding quality prediction device 10 changes the grindstone power value included in the equipment data acquired from the grinding machine 1, and the roughness of the camshaft 3 after grinding, which is the quality data. Get the quality value and. Then, the grinding quality prediction device 10 extracts the grindstone power value at the time of grinding the base circle portion from the grindstone power value. After that, the grinding quality prediction device 10 executes machine learning using the grindstone power value of the base circle portion and the corresponding roughness quality value as learning data to generate a quality prediction model.

In the prediction phase, the grinding quality prediction device 10 extracts the grindstone power value of the base circle portion, which is the same condition as at the time of learning, from the grindstone power values acquired from the grinding machine 1. Then, the grinding quality prediction device 10 inputs the extracted grindstone power value into the trained quality prediction model. After that, the grinding quality prediction device 10 acquires the roughness quality prediction value as the output result of the learned quality prediction model.

As described above, in the grinding quality prediction device 10 according to the first embodiment, the transition of the power value of each grindstone when grinding with the grinding wheel 2 is performed on each of the camshafts 3 of various types, and the transition of each of the camshafts 3 Obtain information and information indicating grinding quality. Then, the grinding quality prediction device 10 extracts a partial power value which is a power value when the common portion of each camshaft 3 is ground from the transition of each grindstone power value. After that, the grinding quality prediction device 10 generates a quality prediction model for predicting the grinding state by using the learning data based on the partial power value and the information indicating the grinding quality, so that the prediction accuracy of the grinding quality is improved. be able to.

Further, since the user can grasp the state of the grinding wheel 2 based on the prediction result of the quality prediction model, the grinding wheel 2 can be dressed at an appropriate timing.

[Functional configuration]
FIG. 3 is a functional block diagram showing a functional configuration of the grinding quality prediction device 10 according to the first embodiment. As shown in FIG. 3, the grinding quality prediction device 10 includes a communication unit 11, a storage unit 12, and a control unit 20.

The communication unit 11 is a processing unit that controls communication with other devices, such as a communication interface. For example, the communication unit 11 receives various data related to the power value from the grinding machine 1 and outputs the prediction result to a terminal used by a user such as an administrator.

The storage unit 12 is an example of a storage device that stores various data, a program executed by the control unit 20, and the like, and is, for example, a memory or a hard disk. The storage unit 12 stores grinding data 13, learning data 14, prediction target data 15, and prediction result 16.

The grinding data 13 stores the equipment data when the grinding wheel 2 is driven and the quality data of the camshaft 3 generated by the grinding at that time in association with each other. Specifically, the grinding data 13 is equipment data and quality data for each camshaft for which grinding has been completed. FIG. 4 is a diagram illustrating an example of grinding data 13. As shown in FIG. 4, the grinding data 13 stores the power value as equipment data and the roughness quality value as quality data in association with each other.

The power value stored here is the transition data of the power value in a series of grinding processes when the camshaft 3 is ground, and the roughness quality value is the arithmetic mean roughness (Ra) of the camshaft 3 after the grinding is completed. Value). In the example of FIG. 4, it is shown that the quality of the camshaft 3 at the completion of grinding was "R value", and the transition of the power value at that time was "x0, x1, x2 ...". The power value can be obtained by monitoring the motor or the like of the grinding machine 1, and the roughness quality value is set to a value measured by the user.

Next, the transition data of the power value set in the equipment data will be described. FIG. 5 is a diagram for explaining transition data of the power value. As shown in FIG. 5, the transition data of the power value is data showing the time transition of the power value output for driving and rotating the grinding wheel 2 from the start of grinding to the end of grinding. The vertical axis of FIG. 5 is the power value (A: ampere), and the horizontal axis is time.

The training data 14 is data used for learning a quality prediction model, and is a plurality of data composed of explanatory variables and objective variables. FIG. 6 is a diagram illustrating an example of data 14 for learning. As shown in FIG. 6, the learning data 14 is data in which "explanatory variables (power values) and objective variables (roughness quality values)" are associated with each other. Here, the explanatory variable is the data extracted from the equipment data of the grinding data 13, and the objective variable is the data extracted from the quality data of the grinding data 13. In the example of FIG. 6, the data for learning is shown in which the explanatory variable is (x, y, z ...) And the objective variable is (R).

The prediction target data 15 is data used for quality prediction. Specifically, the prediction target data 15 is a power value when the base circle of the camshaft 3 which is the grinding wheel 2 whose state is to be predicted, which is extracted by the prediction processing unit 40 described later, is ground. The prediction target data 15 is input to the trained quality prediction target model.

The prediction result 16 is an output result from the trained quality prediction target model obtained by inputting the prediction target data 15 into the trained quality prediction target model. For example, the prediction result 16 corresponds to the arithmetic mean roughness (Ra value) or the like.

The control unit 20 is a processing unit that controls the entire grinding quality prediction device 10, such as a processor. The control unit 20 has a learning processing unit 30 and a prediction processing unit 40. The learning processing unit 30 and the prediction processing unit 40 are examples of electronic circuits included in the processor and examples of processes executed by the processor.

The learning processing unit 30 has a grinding result acquisition unit 31, a data generation unit 32, and a learning unit 33, and is a processing unit that generates data for learning and learns a quality prediction model.

The grinding result acquisition unit 31 is a processing unit that acquires information indicating the grinding quality of each camshaft 3 from the grinding machine 1 and the like. For example, the grinding result acquisition unit 31 monitors the motor in which the grinding machine 1 drives the grinding wheel 2 from the start to the end of the grinding of the camshaft 3, acquires the power value generated during grinding, and obtains the power value of the power value. Generate transition data. Then, the grinding result acquisition unit 31 acquires the Ra value of the camshaft 3 at that time by input from the user or the like, associates the transition data of the power value with the Ra value, and stores it in the storage unit 12 as the grinding data 13. .. In this way, the grinding result acquisition unit 31 generates grinding data 13 each time the grinding of the camshaft 3 is completed.

Although an example in which the grinding result acquisition unit 31 generates grinding data has been described here, the present invention is not limited to this. For example, a user such as an administrator can generate an association between the transition data of the power value and the Ra value, and use the data as the grinding data 13.

The data generation unit 32 is a processing unit that extracts the feature amount to be learned from the grinding data 13, generates the learning data 14, and stores it in the storage unit 12. Specifically, the data generation unit 32 specifies the timing of grinding the base circle from the grinding data 13, and generates learning data 14 using the power value of the timing. Further, since the grinding process by the grinding machine 1 executes the grinding of the camshaft 3 through various processes, the learning accuracy can be improved by using the power value at the time of a specific process. Therefore, here, the identification at the time of base circle grinding, the identification of the process, and the generation of data for learning will be described in order.

(1. Identification at the time of base circular grinding)
First, the specification at the time of base circular grinding will be described with reference to FIG. 7. FIG. 7 is a diagram for explaining the specification of the timing for grinding the base circle. By acquiring the coordinates (cutting coordinates) of the grinding wheel 2 from the grinding machine 1, the data generation unit 32 can specify the timing of cutting into the camshaft 3 in order for the grinding wheel 2 to grind.

In the grinding process, when the grinding process is started, the camshaft 3 also starts to rotate. Then, the grinding wheel 2 executes grinding on the base circle when the base circle of the camshaft 3 reaches the grinding position, and grinds other than the base circle when the base circle deviates from the grinding position.

For example, as shown in FIG. 7, when the grinding process is started at time t0, the grinding wheel 2 starts grinding from the top portion of the camshaft 3 in contact with the grinding wheel 2, and the base circle of the camshaft 3 is ground at time t5. When the position is reached, the base circle is ground. Then, the grinding wheel 2 grinds the base circle of the camshaft 3 until the time t8 when it is located at the grinding position, and then grinds the portion other than the base circle. That is, the grinding wheel 2 grinds the base circle of the camshaft 3 only from the time t5 to the time t8.

Here, assuming that the position where the grinding wheel 2 grinds the top portion of the camshaft 3 at the start of grinding is used as a reference, the grindstone cutting axis coordinates indicating the coordinates of the grinding wheel 2 are lower than the reference position (x-axis, y). The values of the axis and z-axis are small). That is, the position is low from the time t5 to the time t8 when grinding is performed on the base circle of the camshaft 3. Therefore, as shown in FIG. 7, the grindstone cutting axis coordinates (position of the grinding wheel 2) become a small value (low position) only while the base circle of the camshaft 3 is being ground, and other timings are large. The value (high position) is reached, and this transition is repeated during the grinding process.

Therefore, the data generation unit 32 monitors the grindstone cutting axis coordinates in the grinding process, identifies a cycle in which the grindstone cutting axis coordinates repeat from a large value to a small value, and bases the small position (low position) within the cycle. Identify as the grinding timing of the circle.

Here, as an example, a position where the reference position is higher than the camshaft 3 has been described as an example, but in the case of the reverse positional relationship, the above period is also reversed, so that the grinding timing of the base circle is also high. It becomes the position.

(2. Identification of each process)
Next, the identification of each process generated in the grinding process will be described with reference to FIG. Generally, each process generated in the grinding process is divided into an air-laboratory process, a rough-laboratory process, a finishing process, and a spark-out process, and the processes transition in this order.

For example, the Kuken process is a process of approaching an object to be ground, and is a process of not grinding the camshaft 3. The rough grinding process is the next process of the empty grinding process, which is the process of performing an overall rough grinding towards the desired quality. The finishing process is the next step of the rough grinding process, is a step of precisely performing grinding toward the target quality, and has each step such as a fine grinding process and a fine grinding process. Grinding in each state of the finishing process has a great influence on the grinding quality. The spark-out process is a process in which the depth of cut is 0 and strain is removed.

For example, the data generation unit 32 can determine each process from the grinding data 13 and the like based on the depth of cut when the grinding wheel 2 cuts to grind the camshaft 3. FIG. 8 is a diagram illustrating determination of each process. FIG. 8 shows the transition of the grindstone shaft power (power value) from the start to the end of grinding and the transition of the grindstone axis coordinates (grinding grindstone 2 position).

As shown in FIG. 8, the data generation unit 32 specifies the period of the grindstone cutting axis coordinates, and specifies the coordinate value below the threshold value as the base circle grinding timing in each period. In the example of FIG. 8, the data generation unit 32 specifies Q as the base circular grinding timing in the case of the period F. Then, the data generation unit 32 calculates the difference between the minimum value and the maximum value at the base circle grinding timing of each cycle as the cutting amount, and specifies each process based on the size of the cutting amount.

For example, as the initial stage of the process increases, the amount of the grinding wheel 2 that cuts into the camshaft 3 also increases. It should be noted that even during the Kuken process, the camshaft 3 is cut but not ground. By setting the threshold value in consideration of these, each process can be specified based on the size of the cutting amount.

To give a specific example, the data generation unit 32 uses the Kuken process if the depth of cut is equal to or greater than the first threshold value, the rough laboratory process if the depth of cut is less than the first threshold value and greater than or equal to the second threshold value, and less than the second threshold value. If it is 0 or more, it is judged as a finishing process, and if it is 0, it is judged as a spark-out process. The relationship between the threshold values is 1st threshold value> 2nd threshold value> 0, and the value can be arbitrarily set according to the specifications and the like. In addition, by further subdividing the threshold value, it is possible to distinguish between the fine research process and the fine research process included in the finishing process. In that case, it is determined that the fine research cut amount is larger than the fine research cut amount. To do.

In this way, the data generation unit 32 extracts the grinding timing of the base circle from the coordinates of the grinding wheel 2, compares the depth of cut at the grinding timing of the base circle with each threshold value, and specifies the timing of each process. To do. In the example of FIG. 8, the data generation unit 32 performs an air-laboratory process from the start of grinding to time T1, a rough-laboring process from time T1 or time T2, a refined process from time T2 to time T3, and time T3 to time T4. The process is specified as the fine research process, and the period from time T4 to time T5 is specified as the spark-out process. In the first embodiment, the process is abbreviated as in the case of Kuken on the drawing.

(3. Generation of learning data)
Next, the generation of data for learning will be specifically described. Data for learning is generated from the method using the power value when grinding the base circle, the method using only the power value of the finishing process when grinding the base circle, and the power value when grinding the base circle. It can be generated by a method of excluding the power value during the process or a combination of these. Here, each method will be specifically described.

(3-1. Method using the power value at the time of grinding the base circle)
First, with reference to FIG. 9, a method using the power value at the time of grinding the base circle will be described. Here, an example will be described in which the learning accuracy can be improved by extracting the power value at the timing when the specifications of the object to be ground and the grinding conditions are common as the feature amount without depending on the model of the camshaft 3.

FIG. 9 is a diagram illustrating an example of generating data for learning using the power value at the time of grinding the base circle. As shown in FIG. 9, the data generation unit 32 specifies the time of base circular grinding (A1 to An (n is a natural number)) and the timing of each process from the grinding data by the above-mentioned method.

Subsequently, the data generation unit 32 specifies a range of power values (A1', A2', ..., An') corresponding to the time of base circular grinding (A1 to An). Then, the data generation unit 32 generates data for learning using each of the specified power values.

For example, the data generation unit 32 calculates a minimum value 1, an average value 1, and a maximum value 1 from a range of power values A1'corresponding to A1 during base circular grinding, and a range of power values corresponding to A2 during base circular grinding. The minimum value 2, the average value 2, and the maximum value 2 are calculated from A2'. In this way, the data generation unit 32 calculates the minimum value, the average value, and the maximum value corresponding to the range of each power value (A1', A2', ..., An'). After that, the data generation unit 32 arranges the calculated values [minimum value 1, average value 1, maximum value 1, minimum value 2, average value 2, maximum value 2, ..., minimum value n, average value. Data for learning is generated by combining n, the maximum value n] with the quality data [R] associated with the original grinding data.

That is, the data generation unit 32 uses [explanatory variable, objective variable] = [(minimum value 1, mean value 1, maximum value 1, minimum value 2, mean value 2, maximum value 2, ..., minimum value n, The learning data 14 having an average value n, a maximum value n), and R] is generated and stored in the storage unit 12. In this way, the data generation unit 32 can generate the learning data 14 from each of the collected grinding data 13, so that the learning accuracy can be improved without depending on the model.

(3-2. Method using the power value of the finishing process)
Next, a method using the power value of the finishing process will be described. Here, an example will be described in which the quality of the grinding wheel 2 can be accurately determined by extracting the power value of the finishing process, which has a large influence on the grinding quality.

FIG. 10 is a diagram illustrating an example of generating data for learning using the power values of the fine research process and the fine research process. As shown in FIG. 10, the data generation unit 32 identifies the finishing processes (B1, B2, B3) during the base circular grinding by the method described above.

Subsequently, the data generation unit 32 specifies a range of power values (B1', B2', B3') corresponding to the finishing process (B1, B2, B3). Then, the data generation unit 32 generates data for learning using each of the specified power values.

For example, the data generation unit 32 calculates a minimum value 1, an average value 1, and a maximum value 1 from the range B1'of the power value corresponding to B1 during the finishing process. Similarly, the data generation unit 32 calculates the minimum value 2, the average value 2, and the maximum value 2 from the range B2'of the power value corresponding to B2 during the finishing process. Further, the data generation unit 32 calculates the minimum value 3, the average value 3, and the maximum value 3 from the range B3'of the power value corresponding to B3 during the finishing process.

After that, the data generation unit 32 arranges the calculated values [minimum value 1, average value 1, maximum value 1, minimum value 2, average value 2, maximum value 2, minimum value 3, average value 3, maximum value. 3] is combined with the quality data [R] associated with the original grinding data to generate learning data.

That is, the data generation unit 32 uses [explanatory variable, objective variable] = [(minimum value 1, mean value 1, maximum value 1, minimum value 2, mean value 2, maximum value 2, minimum value 3, average value 3, average value 3,). The learning data having the maximum value 3), R] is generated and stored in the storage unit 12 as the learning data 14. In this way, the data generation unit 32 generates learning data 14 from each of the collected grinding data 13, so that the model can be learned by the features of the finishing process that most affect the quality. It can be done, and the learning accuracy can be further improved.

(3-3. Method of excluding power value at the time of Kuken)
Next, a method of excluding the power value at the time of Kuken will be described with reference to FIG. The above-mentioned grindstone shaft power value is a current value (A) for rotating the grindstone. Normally, in the grinding process, control is performed to keep the rotation speed of the grinding wheel 2 during processing constant. Then, when the rotation speed of the grinding wheel 2 is about to drop due to some influence, the control of increasing the power value of the grindstone shaft and returning it to the set rotation speed is executed. That is, the grindstone shaft power value can be considered as a resistance element to the rotational component of the grinding wheel 2.

There are two major factors that change the rotation speed of the grinding wheel 2. These are (1) a resistance element when the grinding wheel 2 and the work come into contact with each other, and (2) a resistance element when the grinding wheel 2 is rotated. It is known that the work quality in the grinding process is the quality of the state of the surface of the grindstone, and the grinding resistance during grinding greatly contributes. However, simply acquiring the grindstone shaft power value during the grinding process includes a resistance element when rotating the grinding wheel 2. Therefore, by acquiring the grindstone shaft power value during the Kuken process and subtracting the grindstone shaft power value during the grinding process from the grindstone shaft power value during grinding, the resistance element when the grinding wheel 2 and the workpiece come into contact with each other is subtracted. Only can be extracted. Therefore, here, an example of guaranteeing the grinding quality by learning the feature amount that greatly contributes to the grinding quality by extracting the grindstone shaft power value excluding the grindstone shaft power value during the Kuken process will be described.

FIG. 11 is a diagram illustrating an example of generating data for learning excluding the power value during the Kuken process. As shown in FIG. 11, the data generation unit 32 specifies the timing of the air-laboratory process and the finishing process (seiken, micro-laboratory) from the grinding data by the above-mentioned method.

Subsequently, the data generation unit 32 specifies a range of power values (C1', ..., Cn') corresponding to the timing (C1 to Cn) during the Kuken process. Similarly, the data generation unit 32 specifies a range of power values (B1', B2', B3') corresponding to the finishing process (B1, B2, B3). Then, the data generation unit 32 generates data for learning using a value obtained by subtracting the power value at the time of the Kuken process from the power value at the time of the finishing process.

For example, the data generation unit 32 calculates the minimum value 1, the average value 1, and the maximum value 1 from the power value range C1', and calculates the minimum value 2, the average value 2, and the maximum value 2 from the power value range C2'. To do. In this way, the data generation unit 32 calculates the minimum value, the average value, and the maximum value corresponding to the range of each power value (C1', ..., Cn'). Subsequently, the data generation unit 32 uses each minimum value, each average value, and each maximum value calculated from the range of power values (C1', ..., Cn') at the time of each Kuken process, and minimizes them. The average value (xa), the average value (ya) of the average value, and the average value (za) of the maximum value are calculated.

Then, the data generation unit 32 selects [minimum value 1, average value 1, maximum value 1, minimum value 2, average value 2, maximum value 2, from the range of power values corresponding to each of B1, B2, and B3 during the finishing process. The minimum value 3, the average value 3, and the maximum value 3] are calculated. After that, the data generation unit 32 uses the power value at the time of the finishing process [minimum value 1, average value 1, maximum value 1, minimum value 2, average value 2, maximum value 2, minimum value 3, average value 3, maximum value 3 ], The power value obtained by subtracting the average value (xa) of the minimum value, the average value (ya) of the average value, and the average value (za) of the maximum value [minimum value 1-xa, average value 1-ya, maximum value 1]. -Za, minimum value 2-xa, mean value 2-ya, maximum value 2-za, minimum value 3-xa, mean value 3-ya, maximum value 3-za] are calculated.

After that, the data generation unit 32 combines the calculated power value and the quality data [R] associated with the original grinding data, and data for learning [(minimum value 1-xa, average value 1-). Ya, maximum value 1-za, minimum value 2-xa, average value 2-ya, maximum value 2-za, minimum value 3-xa, average value 3-ya, maximum value 3-za), R] is generated. ..

In this way, the data generation unit 32 can learn the model after excluding the state of the Kuken process, which has the least influence on the quality, from the state of the finishing process, which has the least influence on the quality, and can improve the learning accuracy. It can be further improved.

(3-4. Combination)
The methods 3-1 to 3-3 described above can be executed independently or can be combined arbitrarily. Further, in the method 3-3, an example of excluding the power time during the air-laboratory process from the power value of the finishing process was explained, but the air-laboratory process is performed from each power time during the base circular grinding obtained in 3-1. It is also possible to exclude the power value of time.

Returning to FIG. 3, the learning unit 33 is a processing unit that generates a quality prediction model using the learning data 14. FIG. 12 is a diagram illustrating the generation of the quality prediction model. As shown in FIG. 12, the learning unit 33 inputs the learning data 14 into the quality prediction model using a neural network or the like, and acquires the output result of the quality prediction model. Then, the learning unit 33 calculates an error between the output result of the quality prediction model and the objective variable of the data for learning, and generates a quality prediction model so as to minimize this error.

For example, the learning unit 33 has data for learning [(minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, mean value 2-ya, maximum value 2-za, minimum). Of the values 3-xa, mean 3-ya, maximum 3-za), R], the explanatory variables (minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, The average value 2-ya, the maximum value 2-za, the minimum value 3-xa, the average value 3-ya, the maximum value 3-za) are input to the quality prediction model, and the output value is acquired. Then, the learning unit 33 learns the quality prediction model so that the error between the objective variable (R) of the learning data and the output value becomes small.

Then, when the learning is completed, the learning unit 33 stores the learned quality prediction model as a learning result in the storage unit 12. The timing for completing the learning can be arbitrarily set, such as when the error becomes equal to or less than the threshold value after the elapse of a predetermined time or after the completion of learning with a predetermined number of learning data.

The prediction processing unit 40 has an acquisition unit 41, an extraction unit 42, and a prediction unit 43, and is a processing unit that predicts the quality of grinding by the grinding machine 1 using a learned quality prediction model.

The acquisition unit 41 is a processing unit that acquires data to be predicted. For example, after the quality prediction model is completed, the acquisition unit 41 acquires grinding data including the power value when it is ground by the grinding machine 1 and outputs it to the extraction unit 42.

The extraction unit 42 is a processing unit that extracts the feature amount to be predicted from the grinding data to be predicted. Specifically, the extraction unit 42 uses the same method as when the data generation unit 32 generates the learning data, and inputs the feature amount (prediction target data 15) from the prediction target grinding data to the quality prediction model. Is generated and stored in the storage unit 12.

For example, when the data generation unit 32 generates learning data by using the above-mentioned 3-1 method, the extraction unit 42 predicts the power value at the time of base circle grinding from the grinding data to be predicted by the same method. It is extracted as the target data 15. Further, when the data generation unit 32 generates learning data by using the above 3-3 method, the extraction unit 42 uses the same method to obtain the power value at the finishing process and the power value at the time of the aerial research from the grinding data to be predicted. The power value of the above is extracted, and the prediction target data 15 excluding the power value at the time of Kuken from the power value at the time of the finishing process is extracted.

The prediction unit 43 is a processing unit that predicts the quality of grinding by the grinding machine 1 by using the prediction target data 15 extracted by the extraction unit 42 and the trained quality prediction target model. For example, the prediction unit 43 acquires a trained quality prediction target model from the storage unit 12. Then, the prediction unit 43 inputs the prediction target data 15 extracted by the extraction unit 42 into the trained quality prediction target model and acquires an output.

After that, the prediction unit 43 stores the output from the quality prediction target model in the storage unit 12 as the prediction result, displays it on a display or the like, or transmits it to a user terminal such as an administrator. In this way, the prediction processing unit 40 repeatedly executes the prediction every time the prediction target data 15 is generated and stored in the storage unit 12.

[Processing flow]
Next, each process executed by the grinding quality prediction device 10 will be described. Here, as an example, the flow of each of the learning process and the prediction process when the method 3-3 is used will be described.

(Flow of learning process)
FIG. 13 is a flowchart showing the flow of the learning process. As shown in FIG. 13, the learning processing unit 30 waits until the start of the learning process is instructed by the user or the like (S101), and when the processing start is instructed (S101: Yes), the grinding data from the storage unit 12 13 is acquired (S102).

Subsequently, the learning processing unit 30 extracts power value data from the grinding data 13 (S103), and specifies the base circular grinding timing and each process from the extracted power value data (S104).

Then, the learning processing unit 30 extracts the power value at the time of base circular grinding in the Kuken process (S105), and calculates the average value of the power values of the Kuken process (S106). Further, the learning processing unit 30 extracts the power value at the time of base circular grinding in the finishing process (S107), and subtracts the average value of the power values of the Kuken process from the power value at the time of base circular grinding in the finishing process. (S108).

After that, the learning processing unit 30 generates learning data in which the power value obtained in S107 and the quality data included in the grinding data are associated with each other (S109). Here, when the generation of learning data is continued (S110: No), S102 and subsequent steps are repeated.

On the other hand, when the generation of the learning data is completed (S110: Yes), the learning processing unit 30 learns the quality prediction model using the generated learning data (S111), and when the learning is completed, the learning result. Is stored in the storage unit 12 (S112).

(Flow of prediction processing)
FIG. 14 is a flowchart showing the flow of the prediction process. As shown in FIG. 14, the prediction processing unit 40 waits until the start of the prediction processing is instructed by the user or the like (S201), and when the processing start is instructed (S201: Yes), the prediction processing unit 40 learns from the storage unit 12. Acquire the completed quality prediction model (S202).

Subsequently, the prediction processing unit 40 acquires the grinding data to be predicted from the storage unit 12 and the grinding machine 1 (S203), and specifies the base circular grinding timing and each process from the grinding data to be predicted (S204). ..

Then, the prediction processing unit 40 extracts the power value at the time of base circular grinding in the Kuken process (S205), and calculates the average value of the power values of the Kuken process (S206). Further, the prediction processing unit 40 extracts the power value at the time of base circular grinding in the finishing process (S207), subtracts the average value of the power value of the Kuken process from the power value of the finishing process, and predicts the data 15. Is generated (S208).

After that, the prediction processing unit 40 inputs the generated prediction target data 15 into the trained quality prediction model (S209), acquires the output result of the quality prediction model, and outputs it (S210).

Then, when the prediction processing unit 40 continues the prediction (S211: No), S203 and subsequent steps are repeated, and when it receives an instruction to end the prediction (S211: Yes), the prediction processing unit ends the prediction processing.

[effect]
As described above, in the grinding quality prediction device 10, the transition of each power value that drives the grinding wheel 2 when grinding with the grinding wheel 2 is performed on each of the plurality of camshafts 3, and the transition of each power value that drives the grinding wheel 2 and the plurality of camshafts 3 Obtain information indicating the grinding quality of each. The grinding quality prediction device 10 extracts the power value, which is the power value when the base circle, which is a common part of the plurality of camshafts 3, is ground, from the transition of each power value. After that, the grinding quality prediction device 10 can generate a quality prediction model for predicting the grinding state by using the data based on the extracted power value and the information indicating the grinding quality.

As a result, the grinding quality prediction device 10 can learn a prediction model that does not depend on the model of the camshaft 3, so that the prediction accuracy of the grinding quality can be improved. Further, since the grinding quality prediction device 10 can predict the quality of the camshaft 3 after grinding, as a result, the state of the grinding wheel 2 obtained by grinding the camshaft 3 can also be predicted. Further, since the state of the grinding wheel 2 can be predicted, it is possible to provide a common index for determining the dress timing without depending on the manager. In addition, since the dressing timing can be appropriately determined, unnecessary dressing can be suppressed and the quality assurance can be upgraded.

Further, since the grinding quality prediction device 10 learns the quality prediction model by using the feature amount excluding the power value of the Kuken process, which has a small influence on grinding, the prediction accuracy of the quality prediction model can be improved. .. Further, since the grinding quality prediction device 10 learns the quality prediction model using the power value of the finishing process, which has a large influence on the quality, as a feature quantity, it is possible to acquire the prediction result capable of accurately determining the quality of the grinding wheel 2. it can.

By the way, in addition to the learning data described in the first embodiment, the prediction accuracy can be improved by incorporating other factors that affect the quality of grinding into the explanatory variables. Therefore, in the second embodiment, an example in which the quality prediction model is trained by increasing the dimension of the data for learning by incorporating a new feature amount into the explanatory variable will be described. Here, an example in which the method 3-3 is adopted will be described.

[Diameter of grinding wheel 2]
For example, the data generation unit 32 can incorporate the diameter of the grinding wheel 2 into an explanatory variable. FIG. 15 is a diagram illustrating an example in which the diameter of the grinding wheel 2 is used for learning data. As shown in FIG. 15, when the diameter of the grinding wheel 2 is large, the area in contact with the camshaft 3 is large, so that the grinding resistance is large. That is, since the diameter of the grinding wheel 2 affects the magnitude of the power value and the magnitude of the power value greatly affects the grinding of the camshaft 3, the diameter of the grinding wheel 2 is used as the feature amount for learning. As a result, the prediction accuracy can be improved.

Therefore, the data generation unit 32 uses the learning data [(minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, mean value 2-) described in 3-3 above. Add the diameter of the grinding wheel to the explanatory variables of ya, maximum value 2-za, minimum value 3-xa, mean value 3-ya, maximum value 3-za), R].

That is, the data generation unit 32 uses the learning data [(minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, mean value 2-ya, maximum value 2-za, The learning of the quality prediction model is executed using the minimum value 3-xa, the average value 3-ya, the maximum value 3-za, the diameter of the grinding wheel 2), and R].

[Information on camshaft 3]
For example, when the diameter of the camshaft 3 is different, or when the work rotation speed during grinding is different, the grinding conditions are different, which affects the information indicating the grinding quality. Therefore, the data generation unit 32 improves the learning accuracy and the prediction accuracy by generating the learning data in which the diameter and the rotation speed of the camshaft 3 at the time of grinding are added. FIG. 16 is a diagram illustrating an example in which the diameter and the number of rotations of the camshaft 3 are used as learning data.

As shown in FIG. 16A, the data generation unit 32 uses the learning data [(minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, mean value 2). -Ya, maximum value 2-za, minimum value 3-xa, mean value 3-ya, maximum value 3-za, diameter of camshaft 3), R] are used to train the quality prediction model.

Further, as shown in FIG. 16B, the data generation unit 32 uses the learning data [(minimum value 1-xa, mean value 1-ya, maximum value 1-za, minimum value 2-xa, average). The quality prediction model is trained using the values 2-ya, maximum 2-za, minimum 3-xa, mean 3-ya, maximum 3-za, camshaft 3 rotation speed), R]. To do.

It is also possible to generate a quality prediction model with high prediction accuracy by generating learning data in which the diameter of the base circle of the camshaft 3 and the rotation speed of the grinding wheel 2 are added to the explanatory variables.

Although the examples of the present invention have been described so far, the present invention may be implemented in various different forms other than the above-described examples.

[Data, numerical values, etc.]
The data example, numerical example, display example, threshold value, etc. used in the above embodiment are merely examples and can be arbitrarily changed. The process subdivision is also an example and can be changed arbitrarily. Further, not only the camshaft 3 but also other workpieces having common parts in a plurality of models can be processed in the same manner.

[Data for learning]
In the above embodiment, an example in which the minimum value, the average value, and the maximum value of the power values at the time of base circle grinding in the corresponding process are used for learning data has been described, but the present invention is not limited to this. For example, any one of the minimum value, the average value, and the maximum value may be used, any two may be used, and a variance value or the like may be used. Further, the diameter and rotation speed of the camshaft 3, the diameter of the grinding wheel 2, and the like can be obtained from the grinding machine 1, or can be obtained by input from the administrator or the like.

[Learning model]
In the above embodiment, various machine learning such as a neural network, a support vector machine, and a logistic regression model can be adopted as the learning model.

[system]
Information including processing procedures, control procedures, specific names, and various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific forms of distribution and integration of each device are not limited to those shown in the figure. That is, all or a part thereof can be functionally or physically distributed / integrated in any unit according to various loads, usage conditions, and the like.

Further, each processing function performed by each device may be realized by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware by wired logic.

[hardware]
FIG. 17 is a diagram illustrating a hardware configuration example. As shown in FIG. 17, the grinding quality prediction device 10 includes a communication device 10a, an HDD (Hard Disk Drive) 10b, a memory 10c, and a processor 10d. Further, the parts shown in FIG. 17 are connected to each other by a bus or the like.

The communication device 10a is a network interface card or the like, and communicates with other devices. The HDD 10b stores a program or DB that operates the function shown in FIG.

The processor 10d reads a program that executes the same processing as each processing unit shown in FIG. 3 from the HDD 10b or the like and expands the program into the memory 10c to operate a process that executes each function described in FIG. 3 or the like. For example, this process performs the same function as each processing unit of the grinding quality prediction device 10. Specifically, the processor 10d reads a program having the same functions as the learning processing unit 30 and the prediction processing unit 40 and the like from the HDD 10b and the like. Then, the processor 10d executes a process of executing the same processing as the learning processing unit 30 and the prediction processing unit 40 and the like.

In this way, the grinding quality prediction device 10 operates as an information processing device that executes the grinding prediction method by reading and executing the program. Further, the grinding quality prediction device 10 can realize the same function as that of the above-described embodiment by reading the program from the recording medium by the medium reading device and executing the read program. The program referred to in the other embodiment is not limited to being executed by the grinding quality prediction device 10. For example, the present invention can be similarly applied when another computer or server executes a program, or when they execute a program in cooperation with each other.

10 Grinding quality prediction device 11 Communication unit 12 Storage unit 13 Grinding data 14 Learning data 15 Prediction target data 16 Prediction result 20 Control unit 30 Learning processing unit 31 Grinding result acquisition unit 32 Data generation unit 33 Learning unit 40 Prediction processing unit 41 Acquisition unit 42 Extraction unit 43 Prediction unit

Claims (10)

The computer
The transition of the power value that drives the grinding wheel and the information indicating the quality of the grinding object when the grinding object is ground using the grinding wheel are acquired.
From the transition of the power value, the power value of the common part, which is the power value when the grinding wheel grinds the common part of a plurality of grinding objects to be ground, is extracted.
A grinding condition monitoring method, characterized in that a process of generating a quality prediction model for predicting a grinding condition is executed using data based on a power value of the common portion and information indicating the quality of the grinding object. From the transition of the power value, among the processes from the start of grinding to the end of grinding, the Kuken process in which the grinding wheel does not grind the object to be ground is identified.
The computer executes the process of calculating the excluded power value excluding the power value at the time of the Kuken process from the power value of the common part.
The grinding state according to claim 1, wherein the generated process uses data based on the excluded power value and information indicating the quality of the grinding object to generate the quality prediction model. Monitoring method. From the transition of the power value, among each process from the start of grinding to the end of grinding, the finishing process for finishing the grinding is specified.
The computer executes a process of extracting the power value at the time of finishing, which is the power value at the time of the finishing process, from the power values of the common portion.
The grinding state according to claim 1, wherein the generated process uses data based on the power value at the time of finishing and information indicating the quality of the grinding object to generate the quality prediction model. Monitoring method. The generated process uses the data in which the minimum value, the maximum value, and the average value of the power values of the common portion are the objective variables and the information indicating the quality of the grinding object is the explanatory variable, and the quality prediction model is used. The grinding state monitoring method according to claim 1, wherein the grinding state is generated. The generated process uses the data in which the power value of the common portion and the diameter of the grinding wheel when grinding are performed as the objective variable and the information indicating the quality of the object to be ground as the explanatory variable, and the quality of the grinding wheel is used. The grinding state monitoring method according to claim 1, wherein a prediction model is generated. The object to be ground is a camshaft.
The grinding state monitoring according to claim 1, wherein the extraction process extracts the power value when the base circle of the camshaft is ground as the power value of the common portion from the transition of the power value. Method. The generated process uses the data in which the power value of the common portion and the diameter of the object to be ground when ground by the grinding wheel are objective variables and the information indicating the quality of the object to be ground is an explanatory variable. The grinding state monitoring method according to claim 1, wherein the quality prediction model is generated. The generated process uses the data in which the power value of the common portion and the number of rotations of the grinding object when ground by the grinding wheel are objective variables and the information indicating the quality of the grinding object is an explanatory variable. The grinding state monitoring method according to claim 1, wherein the quality prediction model is generated by using the method. On the computer
The transition of the power value that drives the grinding wheel and the information indicating the quality of the grinding object when the grinding object is ground using the grinding wheel are acquired.
From the transition of the power value, the power value of the common part, which is the power value when the grinding wheel grinds the common part of a plurality of grinding objects to be ground, is extracted.
A grinding condition monitoring program characterized in that a process of generating a quality prediction model for predicting a grinding condition is executed using data based on the power value of the common portion and information indicating the quality of the grinding object. A grinding machine that grinds objects to be ground using a grinding wheel,
An acquisition unit that acquires information indicating the transition of the power value that drives the grinding wheel and the quality of the object to be ground from the grinding machine.
An extraction unit that extracts the power value of the common portion, which is the power value when the grinding wheel grinds the common portion of a plurality of grinding objects to be ground, from the transition of the power value.
A device characterized by having a learning unit that generates a quality prediction model that predicts a grinding state by using data based on a power value of the common portion and information indicating the quality of the grinding object.

Images