A real-time monitoring method for the waste falling state of crop shears based on sound

Abstract

The present invention discloses a real-time monitoring method for the waste dropping state of a crop shear based on sound, and the present invention comprises the following steps: obtaining the sound signal of the on-site shearing process in real time; marking and intercepting the obtained sound signal according to a set window time length; extracting the features of the currently intercepted sound signal; comparing the extracted features with the trained waste dropping sound feature template; if the feature correlation is higher than the set threshold, it is judged that the drop is detected in the current time period; continuously marking and intercepting the next sound signal according to the set window sliding length, processing in parallel, repeating the above-mentioned feature extraction and correlation comparison steps, and detecting whether there is a drop in the new time period. The present invention also includes the steps of predicting the drop and determining the drop state in combination with the on-site situation. The present invention can quickly and effectively detect whether the steel plate waste drops normally after the crop shear is sheared without stopping the operation of the crop shear, and has the advantages of not contacting the operating equipment and convenient and efficient testing.

Description

Technical Field

[0001] The invention relates to the field of monitoring the operating state of a shearing process of a wide and thick plate cropping shear, and in particular to a method for judging the falling state of waste materials after shearing by using sound. Background Art

[0002] Wide and thick plates are widely used in construction engineering, machinery manufacturing, container manufacturing, shipbuilding, bridge construction and other fields. In the production and processing steps, the head shear is an indispensable process. The head shear is mainly used to cut off the irregular head and tail of the steel plate and to segment the steel plate to prepare for the subsequent double-sided shearing of the edges. The cut head (tail) slides from the waste chute at the bottom of the shear to the waste conveyor belt, and is transported to the waste processing span through the waste conveyor belt. Due to the irregular head and tail and the difficulty in calculating the shearing length accurately, the waste will be stuck in the bottom of the shear after shearing and cannot slide smoothly to the waste conveyor belt. If not discovered in time, it may cause damage to the shearing equipment.

[0003] At present, the method to judge whether the waste has fallen smoothly is to manually check the monitoring screen after shearing. However, since the camera is installed near the outlet of the waste chute at the bottom of the shearing machine, the light is insufficient, and the waste is similar in color to the background environment, it is difficult to identify from the monitoring screen, and the labor intensity is high. Through the sound on the scene, the operator can hear the waste falling, but the operator's monitoring position will be limited, and the noise on the scene is harmful to the human body. Especially under the trend of smart manufacturing and smart factories, it is unreasonable, unscientific and unintelligent for centralized control factories to arrange personnel to monitor on the scene. In addition, relying on manual monitoring is prone to lack of concentration and untimely response. Summary of the invention

[0004] Aiming at the problem that the state of waste falling during the shearing process of the crop shear relies on manual judgment, the present invention provides a real-time monitoring method for the state of waste falling of the crop shear based on sound. The method can quickly identify the state of waste falling after shearing during the operation of the crop shear, without contacting the equipment or requiring manual monitoring, thus ensuring the safety of the equipment and improving the automation of the shearing process.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A sound-based real-time monitoring method for the waste falling state of a crop shear, comprising:

[0007] (1) Real-time acquisition of sound signals of the on-site cutting process: by arranging network microphones at the on-site monitoring points, the sound signals of the on-site cutting process are collected in real time and transmitted to the network recording host computer for storage;

[0008] (2) Pre-judgment of drop: determine in advance whether waste will drop during the shearing action. If it is determined that waste will drop during shearing, proceed to the subsequent steps;

[0009] (3) Data capture: The acquired sound signal is captured according to the set window time length mark;

[0010] (4) Feature extraction: extracting the features of the currently captured sound signal;

[0011] (5) Dropping recognition: The features extracted in step (4) are compared with the trained waste dropping sound feature template. If the feature correlation is higher than the set threshold, it is determined that a drop is detected in the current time period;

[0012] (6) Drop status determination: Under the corresponding shear type, if the drop is recognized within the set time, the drop status is normal. If the drop is still not recognized after the set time, the drop status is abnormal.

[0013] Furthermore, in step (1), the sound signal of the on-site cutting process is obtained in real time by arranging a network microphone on one side of the head cutting shear to pick up the sound signal of the on-site cutting process, and transmit it to the network recording host in real time through the network to achieve storage, playback and analysis on the computer side.

[0014] Furthermore, in step (2), the specific method for predicting the drop is as follows: the type of this shearing is known in advance according to the existing real-time shearing production schedule. If the shearing type is head cutting or tail cutting, it is determined that there should be waste falling after this shearing; if the shearing type is segmentation, there will be no waste falling after shearing, and there is no need to enable subsequent calculations.

[0015] Furthermore, in step (3), intercepting the acquired sound signal according to the set window time length mark specifically refers to:

[0016] After the corresponding monitoring signal of cutting head or tail appears, start the window, intercept the transmitted sound signal according to the set initial window time length, and mark the reference start and end time and physical start and end time of the current sound; wait for a window sliding time length, intercept and mark a new sound signal; continue to intercept and mark new sound signals until the judgment of this cutting drop state is completed;

[0017] Among them, the monitoring signal corresponding to the cutting head is the action signal of the shearing blade, and the monitoring signal corresponding to the cutting tail is the pushing signal of the pushing tail device.

[0018] In this step, data capture focuses on the time period when the drop is most likely to occur. By sliding the window method, the sound data of the new window is captured every other window sliding time length, and the start and end time of each sound segment are marked. The window sliding time is set to 0.1s and the window time length is 1s. The start and stop of the data capture step are controlled by the drop status determination step.

[0019] Furthermore, in step (4), extracting the features of the currently intercepted sound signal means: extracting the multidimensional Mel-frequency cepstrum coefficients of the sound, taking the mean value of the extracted two-dimensional feature matrix in the frame dimension, and forming a one-dimensional feature vector of the sound signal. Specifically, the sound data obtained in step (3) is subjected to the steps of framing, windowing, FFT, mel filter bank filtering, logarithmic operation, and DFT to extract its multidimensional Mel-frequency cepstrum coefficients.

[0020] Furthermore, in step (5), the method for generating the trained waste falling sound feature template is:

[0021] Obtain a large amount of historical sound signals, and extract the sound data containing waste dropping within a set window time length (1s) from the historical sound signals. According to the method of step (4), extract the sound features of each sample, take the average of the feature vectors of all samples, and form a waste dropping sound feature template.

[0022] Furthermore, in step (5), the extracted features are compared with the trained waste falling sound feature template, specifically:

[0023] Calculate the correlation coefficient ρ between the one-dimensional feature vector of the sound signal extracted in step (4) and the feature template of the waste falling sound XY :

[0024]

[0025] X is the one-dimensional feature vector of the sound signal extracted in step (4), Y is the trained waste drop sound feature template vector, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, and E(X) and E(Y) are the expectations of X and Y respectively.

[0026] If the calculated correlation coefficient is greater than the set threshold value of 0.9, it is determined that a drop is detected in the current time period.

[0027] Furthermore, in step (3), according to the sliding time window method, new sound signals are continuously intercepted and obtained. After each window sliding time, the new sound signal obtained in step (2) is calculated in steps (4) to (6). The calculation of each sound segment is performed in parallel without waiting for the processing and judgment of the previous signal to be completed; in this way, the occurrence of a drop can be quickly identified and detected.

[0028] Furthermore, the drop status determination is based on different types of shearing actions, data is intercepted when different signals are received, and a countdown for the corresponding drop action to occur is set, and the drop status result of this shearing action is given according to the result of the drop identification; specifically:

[0029] Get the type of cutting action to be performed from the production plan screen;

[0030] If the shearing action type is head cutting, when the signal of the shear button is captured, the data interception of step (3) is started, the data processing is performed according to step (4), and the countdown T of the head cutting falling is started. head During the countdown, the result of the drop recognition in step (5) is continuously monitored. If a drop is recognized, it is determined that the shear drop is normal. If the drop is not recognized after the countdown, it is determined that the shear drop is abnormal.

[0031] If the tail cutting is a shearing action type, when the tail pushing device action signal is captured, the data capture of step (3) is started, the data processing is performed according to step (4), and the countdown T for the tail cutting to fall is started. tail During the countdown, the result of the drop recognition in step (5) is continuously monitored. If a drop is recognized, it is determined that the shear drop is normal. If the drop is not recognized after the countdown, it is determined that the shear drop is abnormal.

[0032] Specifically, the drop state determination step needs to maintain real-time communication with the original system. Taking each shear as a unit, before the shear action, the shear type is obtained in advance through the HMI, and then different signals are monitored to shorten the monitoring drop time range, start the data interception step, and start to quickly judge the sound signal of each window. Before the set time is reached, if the drop recognition step identifies the drop, it is determined that the drop state of this shear is normal, otherwise it is determined that the drop state of this shear is abnormal, the state determination result appears, and data interception is stopped.

[0033] (1) The shearing type is head cutting, corresponding to the monitoring shear button signal, set the countdown T head =2s; because the type is head cutting, after the shearing action of the shearing machine blade is completed, the waste will fall directly. According to experience, it is set that the waste should fall within 2s.

[0034] (2) The shearing type is tail cutting, corresponding to the action signal of the tail pushing device, set the countdown T tail =5s; because the type is tail cutting, after the shearing action of the shearing machine is completed, the waste remains on the roller. It is necessary to wait for the tail pushing device to act and push the waste forward before it falls. It is set that the waste should fall within 5s.

[0035] Beneficial technical effects of the present invention:

[0036] By monitoring and judging the waste falling state of each shearing action through the method provided by the present invention, the original manual monitoring method is replaced, which can reduce the energy consumption of operators and solve the instability and untimeliness caused by manual monitoring. By combining the on-site action signal, the range of sound monitoring and identification is shortened, unnecessary calculations are reduced, and the accuracy of drop identification is increased. The design of data interception and parallel calculation with the idea of ​​sliding window ensures the timeliness of drop identification. Identifying the falling state as early as possible facilitates the next process action, improves the rhythm of the shearing process, and helps to improve production efficiency. The method provided by the present invention can improve the degree of automation of the shearing process and cater to the development trend of smart factories with fewer people. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0038] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0039] Figure 2 This is a flow chart of the fall prediction step in an embodiment of the present invention;

[0040] Figure 3 This is a flow chart of feature extraction steps in an embodiment of the present invention;

[0041] Figure 4 It is a flow chart of the step of determining the falling state in an embodiment of the present invention - head cutting;

[0042] Figure 5 This is a flowchart of the step of determining the falling state in an embodiment of the present invention - tail cutting;

[0043] Figure 6 This is a sound waveform diagram of the cutting head shearing head falling process in an embodiment of the present invention;

[0044] Figure 7 This is the drop correlation identification result in the embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0047] The embodiment of the present invention provides a method for real-time monitoring of the waste material falling state of a crop shear based on sound. Figure 1 This is a flow chart of the method, including a feature template step trained according to historical waste falling sounds, a real-time sound acquisition step, a falling prediction step, a data interception step, a feature extraction step, a waste falling identification step, and a falling state determination step. Taking each shearing as a unit, firstly, it is predicted whether there should be a fall, and within the corresponding time limit, the falling is monitored using the falling identification step. If within the time limit, the falling prediction and the identification results are consistent, then the falling state is normal. If not, the falling state is abnormal, and an abnormal alarm is issued to prompt the operator to interrupt the equipment action in time to check the situation to avoid equipment damage. The present invention can quickly identify the falling state of waste after shearing during the operation of the head shear, without contacting the equipment, and without manual monitoring, so as to monitor and ensure the safety of the process equipment.

[0048] like Figure 1 As shown, the method comprises the following steps:

[0049] (1) Obtain the sound signal of the on-site shearing process in real time;

[0050] (2) Pre-judgment step of dropping: determine in advance whether waste should drop during each shearing action. If it is determined that waste should drop during shearing, proceed to the subsequent steps;

[0051] (3) Data interception step: intercepting the acquired sound signal according to the set window time length mark;

[0052] (4) Feature extraction step: extracting the features of the currently intercepted sound signal;

[0053] (5) Dropping recognition step: the extracted features are compared with the trained waste dropping sound feature template. If the feature correlation is higher than the set threshold, it is determined that a drop is detected in the current time period;

[0054] (6) Drop status determination step: Under the corresponding shear type, if the drop is recognized within the set time, the drop status is normal. If the drop is still not recognized after the time is up, the drop status is abnormal.

[0055] In this embodiment, step (1) collects the sound signal of the on-site shearing process in real time by arranging a network microphone at the on-site monitoring point, and transmits it to the network recording host computer for storage. Specifically, the network microphone can be arranged on the fixed side of the head shear body, that is, on the right side of the logistics direction, and the microphone sampling rate is selected to be 16KHz. Only the data of the left channel is analyzed, stored in wav format, and transmitted to the computer through the network.

[0056] In this embodiment, step (2) is as follows Figure 2 As shown, the type of shearing action to be performed is obtained from the real-time production plan of the original shearing control system. If it is a head or tail cut, the shearing should have a drop, recorded as PreDropA = 1, otherwise the shearing should not have a drop, recorded as PreDropA = 0. Only the state of the shearing action with a drop is subsequently monitored.

[0057] In this embodiment, step (3) needs to be combined with the monitoring situation of the falling state determination step, such as Figure 4 If the current cutting type is head cutting, when the cutting button signal is detected, the data interception step of this step is started, and the sound data transmitted by the interception step (1) is collected. According to the sliding window method, a series of sound signals are intercepted and marked. The window time length is set to 1s, the sliding time length is set to 0.1s, and the reference start and end time of each sound signal is marked as 0-1s, 0.1-1.1s, 0.2-1.2s, 0.3-1.3s... The physical start and end time is recorded according to Beijing time year, month, day, hour, minute, and second. If the cutting type is tail cutting, when the action signal of the tail pushing device is detected, the data interception step of this step is started. Each time a sound signal is obtained, it is sent to the subsequent steps for feature extraction, correlation comparison and other calculations.

[0058] In this embodiment, step (4), the feature extraction step, specifically performs the following steps on the sound data obtained in step (3): Figure 3 The feature extraction calculation steps shown in the figure first extract the multi-dimensional Mel frequency cepstrum coefficients, and then take the average operation on the frame dimension of the extracted two-dimensional feature matrix to form a one-dimensional feature vector of the sound signal. Among them, the frame length is set to 256 points, the frame shift is 80; the window is a Hamming window; the order of the mel filter is 24, the length of the FFT transform is 256, the sampling frequency is 16KHz, and the DFT discrete cosine transform obtains 12-dimensional Mel frequency cepstrum coefficients. The 12-dimensional Mel frequency cepstrum coefficients of all frames are obtained, and the average value is taken to obtain a 12-dimensional feature vector as the feature of the sound signal.

[0059] In this embodiment, the trained waste falling sound feature template mentioned in step (5) specifically includes the following steps:

[0060] S1: Obtain a large amount of historical sound signals, including the sound of scrap falling and the sound of other actions of the crop shear during the entire shearing process;

[0061] S2: The sound dataset of the cutting process is captured. Each sound segment lasts for 1 second and contains the data of the falling sound, with the label 1.

[0062] S3: Extract common features of the dropped sound data, and use intra-class correlation to measure the quality of the extracted features. Finally, extract the features of the dropped sound data using the feature extraction method described in step (4).

[0063] S4: Take the average of the features extracted from all the dropping sound data in S3 as the waste dropping sound feature template.

[0064] The feature vector obtained in step (4) is compared with the trained waste drop sound feature template vector, and the correlation is calculated according to the following formula:

[0065]

[0066] X is the one-dimensional feature vector of the sound signal extracted in step (4), Y is the trained waste drop sound feature template vector, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, E(X) and E(Y) are the expectations of X and Y respectively, ρ XY is the calculated correlation coefficient. If the calculated correlation coefficient is greater than the set threshold value 0.9, it is determined that a drop is detected in the current time period.

[0067] In this embodiment, each time a new sound data is generated in step (3), it is sent to the subsequent steps (4), (5), and (6) for calculation without waiting for the analysis of the previous sound data to be completed.

[0068] In this embodiment, step (6) is a step of determining the falling state, combined with Figure 4 , 5 Under the corresponding shear type, if the drop is recognized within the set time, the drop state is determined to be normal. If the drop is not recognized after the time is up, the drop state is abnormal.

[0069] like Figure 6 This is a waveform diagram of the shearing process of a steel plant, in which waste material falls during the period of 12.5-13.5s. Figure 7 The method according to the present invention obtains Figure 6Judgment of drop correlation in the cutting process of the head shear. Take the sound with a duration of 1s as the minimum recognition object, and perform an interception recognition calculation every 0.1s of the sliding window duration. In the figure, the horizontal axis is time, the vertical axis is the calculation result of the correlation between each sound segment and the drop template, and the black curve is the drop correlation obtained from the end time of each sound segment. If the correlation is above 0.9, it is determined that a drop has been detected. The first drop is recognized at 13.1s, at which time a drop has indeed occurred, and the results of other action sound recognition in the whole cutting process are lower than 0.9. It can be concluded that this method is fast and feasible.

Claims

1. A sound-based real-time monitoring method for the waste falling state of a crop shear, characterized in that: The method comprises the following steps: (1) Real-time acquisition of sound signals of the on-site shearing process: by arranging network microphones at on-site monitoring points, real-time acquisition of sound signals of the on-site shearing process; (2) Prediction of drop: Determine in advance whether waste will drop during the shearing action. If it is determined that waste will drop during the shearing action, proceed to the subsequent steps. (3) Data capture: The acquired sound signal is captured according to the set window time length mark; (4) Feature extraction: extracting the features of the currently captured sound signal; (5) Drop recognition: Compare the features extracted in step (4) with the trained waste drop sound feature template. If the feature correlation is higher than the set threshold, it is determined that a drop has been detected in the current time period. (6) Drop status determination: Under the corresponding shear type, if the drop is identified within the set time, the drop status is normal; if the drop is still not identified after the set time, the drop status is abnormal; In step (3), the acquired sound signal is intercepted according to the set window time length mark, specifically: After the corresponding monitoring signal of cutting head or cutting tail appears, the window is started, and the transmitted sound signal is intercepted according to the set initial window time length, and the marked content is the reference start and end time and physical start and end time of the current sound; Wait for a window sliding time length, intercept and mark a new sound signal; continue to intercept and mark new sound signals until the current cutting and dropping state is determined; Among them, the monitoring signal corresponding to the cutting head is the action signal of the shearing blade, and the monitoring signal corresponding to the cutting tail is the pushing signal of the material pushing tail device; In step (5), the extracted features are compared with the trained waste drop sound feature template, specifically: Calculate the correlation coefficient between the one-dimensional feature vector of the sound signal extracted in step (4) and the feature template of the waste falling sound : X is the one-dimensional feature vector of the sound signal extracted in step (4), Y is the trained waste drop sound feature template vector, Cov (X, Y) is the covariance of X and Y, D (X) and D (Y) are the variances of X and Y respectively, E (X) and E (Y) are the expectations of X and Y respectively, if the calculated correlation coefficient If it is greater than the set threshold of 0.9, it is determined that a drop is detected in the current time period.

2. The method for real-time monitoring of the waste falling state of the crop shears based on sound according to claim 1, characterized in that: In step (2), the specific method of predicting the drop is as follows: the type of this shearing is known in advance according to the existing real-time shearing production schedule. If the shearing type is head cutting or tail cutting, it is determined that waste should fall after this shearing.

3. The method for real-time monitoring of the waste falling state of the crop shears based on sound according to claim 1, characterized in that: In step (4), the features of the currently intercepted sound signal are extracted, specifically: the multi-dimensional Mel-frequency cepstrum coefficients of the sound are extracted, and the extracted two-dimensional feature matrix is ​​averaged in the frame dimension to form a one-dimensional feature vector of the sound signal.

4. The method for real-time monitoring of the waste falling state of the crop shears based on sound according to claim 1, characterized in that: In step (5), the method for generating the trained waste falling sound feature template is: A large amount of historical sound signals are obtained, and sound data containing waste dropping within a set window time length are intercepted from the historical sound signals, and the multidimensional Mel-frequency cepstrum coefficients of each sound sample are extracted. The extracted two-dimensional feature matrix is ​​averaged in the frame dimension to form a one-dimensional feature vector of the waste dropping sound signal, that is, a waste dropping sound feature template is formed.

5. The method for real-time monitoring of the waste falling state of the crop shears based on sound according to claim 1, characterized in that: In step (3), new sound signals are continuously intercepted according to the sliding time window method. After each window sliding time, the new sound signal obtained in step (2) is calculated in parallel with steps (4) to (6). The calculation of each sound segment is performed in parallel without waiting for the processing and judgment of the previous signal to be completed.

6. The method for real-time monitoring of the waste falling state of the crop shears based on sound according to claim 1, characterized in that: The drop status determination is based on different types of shearing actions, data is intercepted when different signals are received, and a countdown is set for the corresponding drop action to occur. According to the result of drop identification, the state result of the drop of this shearing action is given; specifically: Obtain the type of cutting action to be performed from the production plan screen; If the shearing action type is head cutting, when the signal of the shear button is captured, the data interception of step (3) is started, the data processing is performed according to step (4), and the countdown T for the head cutting to fall is started. head During the countdown, the result of the drop recognition in step (5) is continuously monitored. If the drop is recognized, it is determined that the shearing drop is normal; if the drop is not recognized at the end of the countdown, it is determined that the shearing drop is abnormal; if the shearing action type is tail cutting, when the action signal of the tail pushing device is captured, the data interception of step (3) is started, and the data processing is performed according to step (4), and the countdown T of the tail cutting drop is started. tail During the countdown, the result of the drop recognition in step (5) is continuously monitored. If a drop is recognized, it is determined that the shear drop is normal. If the drop is not recognized after the countdown, it is determined that the shear drop is abnormal.

Images