CNN-based data analysis method

Abstract

The invention relates to a CNN network-based data processing method and a machine tool early warning processing system, and the method comprises the following steps: S1, obtaining original data, and collecting the audio information and monitoring video of each region in a factory through a collection device; s2, classifying and screening the data, classifying the obtained data, and enabling the audio information to be in one-to-one correspondence with the monitoring videos; s3, establishing an audio information reference early warning line, and setting a trigger model at the same time; and S4, screening fault segments, and carrying out early warning in advance. By the adoption of the technical scheme, the established audio information reference early warning line can give an alarm before a cutting problem of the cutter occurs, the situation that a machined part is damaged due to the problem of the cutter is avoided, and therefore the machining efficiency is improved, and the rate of finished products is increased.

Description

Technical Field

[0001] The present invention relates to the field of information data processing, and in particular to a data analysis method based on CNN. Background Art

[0002] Enterprise information management belongs to the category of information resource management at the micro level. It refers to the use of modern management methods and means by enterprises to organize, plan, coordinate and control information resources and information activities related to the enterprise in order to achieve the predetermined goals, so as to realize the reasonable development and effective use of enterprise information management. With the overall transparency of enterprise work management, managers and employees are required to effectively implement work plans within the planned time and limited resources to achieve enterprise goals. At present, the management systems of many enterprises rely only on human statistics, which are not intuitive enough and have low management efficiency. In the actual production of machine tools, various materials will emit fixed characteristic sounds of various frequencies at different processing stages. Experienced operators can tell what kind of fault may occur at a glance. If there is an automatic system that can replace manual processing and judgment more quickly, it can reduce the occurrence of faults and avoid more serious accidents. In the process of internal information management of enterprises, the processing information and fault information of each machine tool in the workshop are integrated into the information management system for reference. It has become a trend of intelligent management of enterprises. However, the existing statistics of frame processing information are imperfect. How to use and collect information inside the workshop has become an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a data analysis method based on CNN to solve the problems existing in the above-mentioned prior art.

[0004] The above technical objectives of the present invention are achieved through the following technical solutions:

[0005] A data analysis method based on CNN, comprising the following steps:

[0006] Step S1, obtaining raw data, and collecting audio information and monitoring video of each area inside the factory through a collection device;

[0007] Step S2: data classification and screening, classifying the acquired data and matching the audio information with the surveillance video one by one;

[0008] Step S3, establishing an audio information baseline warning line and setting up a trigger model at the same time;

[0009] Step S4: Screen the faulty fragments and issue early warning.

[0010] By adopting the above technical solution, the established audio information baseline warning line can issue a warning before the tool has a cutting problem, avoiding damage to the processed part due to tool problems, thereby improving processing efficiency and increasing the yield.

[0011] In a further embodiment, the step S3 further comprises:

[0012] Step S31, obtaining three lines for predicting the remaining service life of the cutting tool under normal conditions, and performing weighted adjustment;

[0013] Step S32: Acquire processing data, include the processing data in the early warning model, establish an alarm model, adjust the alarm model to a multi-level alarm, and perform alarm classification according to the risk model;

[0014] Step S33, establish a neural network, set up activation permissions, and determine whether to activate; if so, perform data calculations, input data into the input layer and perform calculations and analysis; if not, do not activate the computing unit.

[0015] In a further embodiment, the weighting method in step S31 is:

[0016] Q%=-(R / (V*L%*(C-20))),

[0017] Where Q is an arbitrary constant, R is the Rockwell hardness value of the material to be cut, V is the volume of the area to be cut, and the unit of V is cubic centimeters, L% is the cutting feed rate during cutting, which represents the percentage of the tool diameter fed each time cutting, and C is the average temperature of the cutting tool.

[0018] In a further embodiment, the neural network establishment method of step S32 is:

[0019] An input layer, a hidden layer and an output layer are set up, wherein the input layer is used to input original audio data, the hidden layer is used to calculate and compare the original audio data with three tool life lines, and the output layer is used to output the predicted value of the remaining tool life in real time.

[0020] In a further embodiment, the activation authority includes activation verification of the data input terminal and an activation function during data operation, wherein the activation verification of the data input terminal is used to manually screen the data that needs to be predicted, and the activation function is used to search for characteristic peaks in the audio data for reverse transmission and recalculation.

[0021] In a further embodiment, the multi-level alarm establishment condition in step S32 is:

[0022] D = WK, where W represents the number of times the life prediction is lower than the safety curve in the last ten minutes of processing time, and K represents the number of times a dangerous band with a fitting degree of more than 80% appears in the last ten minutes of processing time. W belongs to [1, 2, 3, 4, 5], and the value of W is a positive integer from top to bottom.

[0023] In a further embodiment, the warning system includes a level one light warning, a level two sound warning, a level three light warning and sound warning, a level four high-decibel sound warning, and a level five broadcast prompt warning.

[0024] In a further embodiment, the data screening method includes manual screening and intelligent denoising.

[0025] In summary, the present invention has the following beneficial effects:

[0026] 1. The established audio information baseline warning line can issue a warning before the tool has a cutting problem, avoiding damage to the processed part due to tool problems, thereby improving processing efficiency and increasing the yield rate;

[0027] 2. By setting up multi-level alarm conditions, it is possible to perform real-time calculations and analyze alarm levels based on the situation of the workpiece, making it easier for staff to determine whether emergency stops are needed to reduce damage to parts and equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is an overall flow chart of a data analysis method based on CNN of the present invention;

[0029] Figure 2 It is a neural network structure diagram of a CNN-based data analysis method of the present invention;

[0030] Figure 3 It is a schematic diagram of audio data comparison of a CNN-based data analysis method of the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0032] The same parts are denoted by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the attached Figure 1In the description of this specification, the words "bottom" and "top", "inside" and "outside" refer to directions toward or away from a particular component geometry, respectively. In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this specification, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0033] Embodiment 1:

[0034] like Figure 1-Figure 3 As shown, a data analysis method based on CNN includes the following steps:

[0035] Step S1, obtain the original data, collect the audio information and monitoring video of each area inside the factory through the collection device, the original data is collected by multiple video monitoring matrices set up in the factory and the sound collector set inside each high-precision machining center, and each corresponding video monitoring matrix and each sound collector inside the machine tool have corresponding numbers. When the original data is transmitted back to the terminal processor, it needs to be cut and packaged according to the time period T, and each data packet has a corresponding code, and the coding format is numbered as device model-device number-time information. In order to further identify the data packet and save computing power, it is necessary to manually give an initial set weight when each data packet is sent. The initial weight is 1-5. In actual operation, the initial weight is set for each machine. The weight information of the data packet is matched according to the weight of the machine. The larger the weight vertically, the higher the importance of the data. In order to facilitate production and experiments, the weights of all experimental equipment are set to 5. When transmitting, the data packets with a weight value of 5 are uniformly sent first;

[0036] Step S2, data classification and screening, classify the obtained data, and correspond the audio information to the monitoring video one by one. In order to avoid transmission interruption, the present invention needs to separate the wired data transmission line and the wireless data transmission line into multiple channels. The main frequency channel is 80MHz-120MHz. The frequency band is divided according to the number of data packets with a weight of 5 detected, and the data packets with a unified number are recorded as 1 data packet with a weight of 5. When a new data packet with a weight of 5 appears, the channel is reallocated. When the number of data packets with a weight of 5 is greater than 6 at the same time, a data delay alarm is issued, and at the same time, the auxiliary frequency of 125MHz-160MHz channels is started to perform a new allocation mechanism. The data packet with a weight of 5 is only used to mark and transmit the data packets of the experimental machine and the equipment that is about to have problems. The experimental machine is suitable for timely data collection for calculation. The weight value of the equipment that is about to have problems is automatically adjusted to 5 in order to timely calculate whether a problem is about to occur, so as to avoid large losses caused by equipment damage.

[0037] Step S3, establish an audio information baseline warning line, and set up a trigger model at the same time. The audio information baseline warning line is the baseline service life of tools with different signals of different materials obtained based on historical data. The three baselines are respectively the baselines of the remaining service life of the tool of 2 hours, 6 hours, and 12 hours. The existing tools used for high-precision processing are generally composed of a tool bar and a blade, so it is necessary to perform a test cutting after the blade is installed to detect the installed new tool. The first purpose is to detect whether the installation is correct and qualified. The second purpose is that because the tool bar cannot be directly tested, it is necessary to determine whether the tool bar is in normal use after the blade is installed. If the test cutting data of the new tool is within the range of all the recorded test cutting data, the tool bar is judged to be normal. If it exceeds the range, the tool bar is judged to be abnormal.

[0038] The step S3 further includes: step S31, obtaining three lines for predicting the remaining service life of the cutting tool under normal circumstances, and performing weighted adjustment; the weighted method in the step S31 is: Q%=-(R / (V*L%*(C-20))), where Q is an arbitrary constant with an edge, R is the Rockwell hardness value of the material to be cut, V is the volume of the area to be cut, and the unit of V is cubic centimeters, L% is the cutting feed amount during cutting, representing the percentage of the tool diameter fed each time cutting, C is the average temperature of the cutting tool, and the C value is constantly changing, but because coolant is used during the cutting process, the calculation method is to detect the temperature inside the area through an infrared temperature sensor, and then screen it. Finally, the average temperature is calculated. In actual use, an infrared temperature sensor is installed on the spindle inside the machine tool to monitor the temperature of the tool under the spindle. Because the infrared temperature sensor moves with the position of the spindle, the relative position of the infrared temperature sensor and the tool locked under the spindle does not change. By entering the model of the tool before starting to mark which area is the temperature measurement area of ​​the tool, the temperature detection of the entire cutting area of ​​the tool can be achieved. The model of the tool can be obtained by comparing the temperature measurement data of the infrared temperature sensor with the processing data of the sawing machine. In high-precision automated processing lathes, they are all controlled by the processing program. The model and size of the tool are recorded in the program. The temperature measurement area range data of the tool can be easily obtained through conversion;

[0039] Step S32, obtain processing data, include the processing data in the early warning model, and establish an alarm model, adjust the alarm model to a multi-level alarm, and classify the alarm according to the risk model; the neural network establishment method of step S32 is: establish an input layer, a hidden layer and an output layer, the input layer is used to input the original audio data, the hidden layer is used to calculate and compare the original audio data with the three tool life lines, and the output layer is used to output the tool remaining life prediction value in real time. Assuming that the diameter of the tool is A mm, the feed amount for tool cutting can be 0.1AA. By measuring the feed at 0.3A, which is often said to be 30%, 0.5A, which is often said to be 50%, and 0.7A, which is often said to be 70%, the tool is from the initial The energy value collected by the sound collector during the cutting process from the time the tool is used to the time it is scrapped can be used to obtain the standard service life of the tool in different states under standard data. Then, the characteristic energy waves of the tool when its service life is about to end under standard conditions are collected and marked to determine what problem with the tool causes the energy wave to fluctuate specifically and enter it into the database. In this way, a fault code library for the tool can be obtained, which can be judged by comparison during actual cutting processing. If a fully fitted energy peak or a fitting degree of more than 99% appears during the actual processing of the tool, an alarm is issued, and the tool problem corresponding to the peak is determined to be the level of alarm feature.

[0040] Step S33, establish a neural network, set up activation authority, and determine whether to activate; if so, perform data calculation, input the data into the input layer and calculate and analyze; if not, do not activate the operation unit; the activation authority includes the activation verification of the data input end and the activation function during the data operation process, the activation verification of the data input end is used to manually screen the data that needs to be predicted, and the activation function is used to search for characteristic peaks in the audio data for reverse transmission and recalculation. The multi-level alarm establishment condition in step S32 is: D=WK, where W represents the number of times the life prediction is lower than the safety curve within the last ten minutes of processing time, K represents the number of times a dangerous band with a fitting degree of more than 80% appears within the last ten minutes of processing time, and W belongs to [1,2,3,4,5], and the value of W is a positive integer taken upward.

[0041] Step S4, screening fault fragments and giving early warning, the warning system includes level one light warning, level two sound warning, level three light warning and sound warning, level four high-decibel sound warning and level five broadcast prompt warning, the data screening method includes manual screening and intelligent denoising.

[0042] Because in the actual use process, the environment of the use site and the experimental site is not exactly the same, so it is necessary to obtain audio data during the first standardized processing after the installation is completed, and compare the audio data obtained for the first time with the standard data, and then give the standard data a constant, and adjust the standard data as a whole through the constant. In the process of obtaining data for the first standard processing at the actual use site, all data waves with a contrast ratio below 95% in the data are marked, and it is determined one by one whether they are human factors. If the data with a contrast ratio below 95% occupies more than 20% of the standard data, the laboratory is notified to check the data for compliance and conduct standard measurements of the standard data at the use site. In this way, the data can be continuously optimized. As the scope of use increases, the standard data is gradually enriched and filled, making it more accurate and reliable, and it can also be used for cutting tools. The faults encountered during the processing are collected and counted, and the monitoring of the tool status during the processing of the tool is improved. The data can be quickly processed and analyzed through the CNN algorithm. Because the monitoring of the tool status requires a large amount of computing power, it is necessary to activate the permission to activate the data analysis and comparison system. The method of the present invention can realize visual real-time tracking of the processing status in some high-precision processing processes, avoiding damage to the material or the machine tool body caused by the improper use of the tool, greatly saving the problem of material waste in the field of high-precision processing technology. Compared with some materials and processing equipment themselves, when necessary, the forced button can be directly activated when the specific changing energy wave is found to control the tool to deviate from the cutting path set by the program, so as to achieve the separation of the tool and the workpiece. The design of the deviation path is to make the tool quickly withdraw along the direction of the main axis.

[0043] In the embodiments disclosed in the present invention, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments disclosed in the present invention can be understood according to specific circumstances.

[0044] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A data analysis method based on CNN, characterized in that: The following steps are involved: Step S1, obtaining raw data, and collecting audio information and monitoring video of various areas within the factory through a collection device; Step S2: data classification and screening, classifying the acquired data and matching the audio information with the surveillance video one by one; Step S3, establishing an audio information baseline warning line and setting up a trigger model at the same time; Step S4: Screen the faulty segments and issue early warning.

2. A CNN-based data analysis method according to claim 1, characterized in that: The step S3 further comprises: Step S31, obtaining three lines for predicting the remaining service life of the cutting tool under normal conditions, and performing weighted adjustment; Step S32: Acquire processing data, include the processing data in the early warning model, establish an alarm model, adjust the alarm model to a multi-level alarm, and perform alarm classification according to the risk model; Step S33, establish a neural network, set up activation permissions, and determine whether to activate; if so, perform data calculations, input data into the input layer and perform calculations and analysis; if not, do not activate the computing unit.

3. A CNN-based data analysis method according to claim 2, characterized in that: The weighting method in step S31 is: Q%=-(R / (V*L%*(C-20))), Where Q is an arbitrary constant, R is the Rockwell hardness value of the material to be cut, V is the volume of the area to be cut, and the unit of V is cubic centimeters, L% is the cutting feed rate during cutting, which represents the percentage of the tool diameter fed each time cutting, and C is the average temperature of the cutting tool.

4. A CNN-based data analysis method according to claim 2, characterized in that: The neural network establishment method of step S32 is: An input layer, a hidden layer and an output layer are set up, wherein the input layer is used to input original audio data, the hidden layer is used to calculate and compare the original audio data with three tool life lines, and the output layer is used to output the predicted value of the remaining tool life in real time.

5. A CNN-based data analysis method according to claim 4, characterized in that: The activation authority includes activation verification of the data input terminal and activation function during data operation. The activation verification of the data input terminal is used to manually screen the data that needs to be predicted, and the activation function is used to search for characteristic peaks in the audio data for reverse transmission and recalculation.

6. A CNN-based data analysis method according to claim 2, characterized in that: The multi-level alarm establishment conditions in step S32 are: D = WK, where W represents the number of times the life prediction is lower than the safety curve in the last ten minutes of processing time, and K represents the number of times a dangerous band with a fitting degree of more than 80% appears in the last ten minutes of processing time. W belongs to [1, 2, 3, 4, 5], and the value of W is a positive integer from top to bottom.

7. The CNN-based data analysis method according to claim 1, characterized in that: The early warning system includes a first-level light warning, a second-level sound warning, a third-level light warning and sound warning, a fourth-level high-decibel sound warning and a fifth-level broadcast prompt warning.

8. The CNN-based data analysis method according to claim 1, characterized in that: The data screening method includes manual screening and intelligent denoising.

Images