Aluminum sheet length high-precision automatic online detection system

Abstract

The present application relates to the technical field of automation detection, in particular to a kind of aluminium sheet length high-precision automatic online detection system, comprising: conveying device, for conveying aluminium sheet and realizing speed closed-loop control;Non-contact detection module, for collecting aluminium sheet image and edge profile data;Data processing unit, for calculating dynamic sampling frequency, data fusion algorithm and generating sorting instruction;Sorting execution mechanism, for receiving sorting instruction and sorting unqualified product;Dynamic compensation module, based on real-time linear velocity adjustment laser sampling frequency and edge compensation coefficient, connect conveying device and data processing unit;Self-learning database, for storing historical detection data and training LSTM prediction model, connect data processing unit through Ethernet interface.The beneficial effects of the present application are: can adopt weighted kalman filtering fusion multi-source data output aluminium sheet length value, improve the accuracy of aluminium sheet length detection, can sort unqualified aluminium sheet, improve sorting efficiency.

Description

Technical Field

[0001] This invention relates to the field of automated testing technology, specifically to a high-precision automatic online testing system for aluminum sheet length. Background Technology

[0002] Aluminum sheets are a key raw material in aerospace, automobile manufacturing, electronic equipment, and building decoration. Their length accuracy directly affects the compatibility of subsequent processing. In the production process of aluminum sheets, early detection of out-of-tolerance aluminum sheets allows for machine shutdown and adjustment, avoiding the scrapping of the entire batch of aluminum sheets. Therefore, it is necessary to inspect aluminum sheets.

[0003] In existing technologies, manual measurement is usually used, such as by measuring tape or vernier calipers. This is not only inefficient, but also easily affected by subjective factors, which can lead to misjudgments and makes it difficult to meet the needs of modern industry for large-volume, high-precision testing. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application proposes a high-precision automatic online detection system for aluminum sheet length. This system can use weighted Kalman filtering to fuse multi-source data and output aluminum sheet length values, thereby improving the accuracy of aluminum sheet length detection and enabling the sorting of defective aluminum sheets, thus increasing sorting efficiency.

[0005] The following is the technical solution of the present invention: a high-precision automatic online detection system for aluminum sheet length, comprising:

[0006] A conveying device for transporting aluminum sheets and implementing closed-loop speed control;

[0007] A non-contact detection module is used to acquire images and edge contour data of aluminum sheets;

[0008] The data processing unit is used to calculate the dynamic sampling frequency, data fusion algorithm, and generate sorting instructions;

[0009] The sorting execution mechanism is used to receive sorting instructions and sort out non-conforming products;

[0010] The dynamic compensation module adjusts the laser sampling frequency and edge compensation coefficient based on real-time linear velocity and connects to the transmission device and data processing unit.

[0011] A self-learning database is used to store historical detection data and train LSTM prediction models, and is connected to the data processing unit via an Ethernet interface.

[0012] As a preferred embodiment of the present invention, the conveying device includes a conveyor belt, a drive motor and an encoder, with the encoder output connected to the input of the dynamic compensation module.

[0013] As a preferred embodiment of the present invention, the non-contact detection module consists of several sets of high-precision laser rangefinders and industrial cameras. The laser rangefinders are symmetrically distributed on both sides of the conveyor belt at a 15° tilt angle, and the distance between the measurement points is 1.2 times the nominal width of the aluminum sheet. The data output terminals of the laser rangefinders and industrial cameras are connected to the FPGA chip of the data processing unit.

[0014] As a preferred embodiment of the present invention, the sorting execution mechanism includes a pneumatic push rod and a temporary storage bin, wherein the pneumatic push rod is connected to the industrial control computer of the data processing unit via an RS485 bus.

[0015] A high-precision automatic online detection method for aluminum sheet length includes the following steps:

[0016] S1. Calibrate the non-contact detection module and load parameters;

[0017] S2. Acquire image and edge contour data, and dynamically adjust the sampling frequency based on linear velocity;

[0018] S3. Based on laser data compensation and defect confidence, a weighted Kalman filter is used to fuse multi-source data to output the aluminum sheet length value;

[0019] S4. Based on the aluminum sheet length value and the nominal length of the aluminum sheet, sort out the unqualified aluminum sheets and dynamically adjust the emergency stop acceleration of the conveyor belt.

[0020] As a preferred embodiment of the present invention, in S1, calibration is performed when the laser rangefinder drift ΔP > 0.005 mm, and the calibration expression is as follows:

[0021] P cal =P raw ×(1+k1·ΔT)

[0022] In the above formula, P cal P represents the measured value of the calibrated laser rangefinder. raw The values ​​are the original laser rangefinder measurements, k1 is the temperature compensation coefficient, and ΔT is the change in ambient temperature.

[0023] As a preferred embodiment of the present invention, in S2, the sampling frequency is dynamically adjusted based on the linear velocity, as expressed below:

[0024]

[0025] In the above formula, f s (t) is the dynamically adjusted laser sampling frequency, v(t) is the real-time conveyor belt linear speed, and k2 is the speed adjustment coefficient.

[0026] In a preferred embodiment of the present invention, in step S3, when the absolute difference between the real-time measured width of the aluminum sheet and the nominal width of the aluminum sheet is greater than 2 mm, laser data compensation is performed, as expressed below:

[0027] L corr =L means +k3·(W act -W nom ) 2

[0028] In the above formula, L corr The measured value of the aluminum sheet length after compensation, L means W represents the original laser measurement length value. act W is the width of the aluminum sheet measured in real time. nom k is the nominal width of the aluminum sheet, and k3 is the error correction parameter.

[0029] As a preferred embodiment of the present invention, a weighted Kalman filter is used to fuse multi-source data to output the aluminum sheet length value, as shown in the following expression:

[0030]

[0031] In the above formula, L is the final aluminum sheet length value output by the fusion process. corr C is the measured value of the compensated aluminum sheet length. d For the defect confidence level, L image k is the length value measured visually, and k4 is the weighting coefficient for the laser data.

[0032] As a preferred embodiment of the present invention, in S4, the conveyor belt emergency stop acceleration is dynamically adjusted, as expressed below:

[0033]

[0034] Δt = t2 - t1 - 500 ms

[0035] In the above formula, a new For the adjusted conveyor belt emergency stop acceleration, a old Δt represents the conveyor belt's emergency stop acceleration before adjustment, t2 represents the time delay, t2 represents the actual feedback time after the sorting execution mechanism completes its action, and t1 represents the time for generating the sorting instruction.

[0036] The beneficial effects of this invention are:

[0037] 1. In this invention, the drift of the laser rangefinder can be calibrated, and the aluminum sheet can be detected based on its nominal length and nominal width, thereby improving the flexibility and accuracy of the detection.

[0038] 2. In this invention, laser data compensation can be performed based on the width of the aluminum sheet, and combined with the defect confidence level, a weighted Kalman filter is used to fuse multi-source data to output the aluminum sheet length value, thereby improving the accuracy of aluminum sheet length detection.

[0039] 3. In this invention, unqualified aluminum sheets can be sorted according to the length value and nominal length of the aluminum sheet, and the conveyor belt acceleration can be dynamically adjusted during sorting to reduce sorting delay and improve sorting efficiency.

[0040] 4. In this invention, training can be performed based on ambient temperature and humidity to predict the change in length caused by thermal expansion, thereby reducing the false detection rate in summer production and improving detection accuracy. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the detection system of the present invention. Figure 1 ;

[0042] Figure 2 This is a schematic diagram of the detection system of the present invention. Figure 2 ;

[0043] Figure 3 This is a flowchart illustrating the detection method of the present invention;

[0044] Figure 4 This is a flowchart of the detection method of the present invention;

[0045] In the diagram: 1. Conveying device; 2. Non-contact detection module; 3. Data processing unit; 4. Sorting execution mechanism; 5. Dynamic compensation module; 6. Self-learning database; 101. Drive motor; 102. Encoder; 103. Conveyor belt; 201. Laser rangefinder; 202. Industrial camera; 301. FPGA chip; 302. Industrial computer; 401. Pneumatic push rod; 402. Temporary storage bin. Detailed Implementation

[0046] To make the technical problems solved by the present invention, the technical solutions adopted, and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1:

[0048] like Figure 1 and Figure 2 As shown, a high-precision automatic online detection system for aluminum sheet length includes:

[0049] Conveying device 1 is used to transport aluminum sheets and achieve closed-loop speed control;

[0050] Non-contact detection module 2 is used to acquire aluminum sheet images and edge contour data;

[0051] Data processing unit 3 is used to perform dynamic sampling frequency calculation, data fusion algorithm and generate sorting instructions;

[0052] Sorting execution mechanism 4 is used to receive sorting instructions and sort out non-conforming products;

[0053] The dynamic compensation module 5 adjusts the laser sampling frequency and edge compensation coefficient based on real-time linear velocity and is connected to the transmission device 1 and the data processing unit 3.

[0054] The self-learning database 6 is used to store historical detection data and train LSTM prediction models, and is connected to the data processing unit 3 via an Ethernet interface.

[0055] The MES system interface is used to transmit test results to the factory management system and supports the Modbus-TCP protocol.

[0056] In this embodiment, the conveying device 1 includes a conveyor belt 103, a drive motor 101, and an encoder 102. The drive motor 101 drives the conveyor belt 103 to convey, thereby causing the aluminum sheet on the conveyor belt 103 to move. The encoder 102 feeds back the linear velocity signal to the data processing unit 3 in real time.

[0057] In this embodiment, the non-contact detection module 2 consists of three sets of high-precision laser rangefinders 201 and an industrial camera 202. The laser rangefinders 201 are symmetrically distributed on both sides of the conveyor belt 103 at a 15° tilt angle, and the distance between the measurement points is 1.2 times the nominal width of the aluminum sheet.

[0058] In this embodiment, the data processing unit 3 has a built-in FPGA chip 301 and an industrial control computer 302 to perform dynamic sampling frequency calculation, data fusion algorithm and generate sorting instructions.

[0059] In this embodiment, the sorting execution mechanism 4 includes a pneumatic push rod 401 and a temporary storage bin 402. After receiving the instruction from the industrial control computer 302, the defective products are pushed to the temporary storage bin 402 by the pneumatic push rod 401 within 500ms.

[0060] In this embodiment, the dynamic compensation module 5 adjusts the laser sampling frequency and edge compensation coefficient according to the real-time linear velocity, and connects the encoder 102 of the transmission device 1 with the FPGA chip 301 of the data processing unit 3.

[0061] In this embodiment, the output of the encoder 102 of the conveying device 1 is connected to the input of the dynamic compensation module 5; the data outputs of the laser rangefinder 201 and the industrial camera 202 are connected to the FPGA chip 301; the industrial computer 302 controls the pneumatic push rod 401 of the sorting execution mechanism 4 via an RS485 bus; and the self-learning database 6 is connected to the industrial computer 302 via a gigabit Ethernet.

[0062] Implementation Plan: When the aluminum sheet on the conveyor 1 enters the detection area of ​​the non-contact detection module 2, the encoder 102 provides real-time feedback of the linear velocity to the dynamic compensation module 5. The dynamic compensation module 5 adjusts the sampling frequency of the laser rangefinder 201. The industrial camera 202 synchronously triggers image capture, and the image data is processed by a convolutional neural network to extract surface defect features. The FPGA chip 301 performs edge compensation calculations on the laser data, and the industrial control computer 302 uses a weighted Kalman filter to fuse multi-source data and outputs the length detection result. If the deviation exceeds ±0.05mm, the sorting execution mechanism 4 is immediately triggered. After receiving the instruction from the industrial control computer 302, within 500ms, the non-conforming product is pushed to the temporary storage bin 402 via the pneumatic push rod 401.

[0063] Example 2:

[0064] like Figure 3 and Figure 4 As shown, a high-precision automatic online detection method for aluminum sheet length includes the following steps:

[0065] S1. Calibrate the non-contact detection module 2 and load parameters;

[0066] S2. Acquire image and edge contour data, and dynamically adjust the sampling frequency based on linear velocity;

[0067] S3. Based on laser data compensation and defect confidence, a weighted Kalman filter is used to fuse multi-source data to output the aluminum sheet length value;

[0068] S4. Based on the aluminum sheet length value and the nominal length of the aluminum sheet, sort out the unqualified aluminum sheets and dynamically adjust the emergency stop acceleration of the conveyor belt 103.

[0069] S5. Store the detection data to the self-learning database 6, train and predict the amount of length change caused by thermal expansion.

[0070] In step S1, the system is initialized and parameters are loaded. Specifically, the system is initialized, and the sensor zero-point drift is checked. If the drift of the laser rangefinder 201 ΔP > 0.005 mm, calibration is performed. The calibration expression is as follows:

[0071] P cal =P raw ×(1+k1·ΔT)

[0072] In the above formula, P cal For the measured values ​​of the calibrated laser rangefinder 201, P raw The values ​​are the original measurements from the laser rangefinder 201, where k1 is the temperature compensation coefficient and ΔT is the change in ambient temperature. In this embodiment, k1 = 0.0015.

[0073] Loading parameters, including the nominal length L of the aluminum sheet. nom Nominal width Wnom And the temperature compensation coefficient k1, wherein the temperature compensation coefficient k1 is obtained through the self-learning database 6;

[0074] In step S2, dynamic sampling and data acquisition take place. Specifically, the sampling frequency is dynamically adjusted based on the linear velocity. The encoder 102 acquires the linear velocity v(t) in real time and dynamically adjusts the sampling frequency, as shown in the following expression:

[0075]

[0076] In the above formula, f s v(t) is the dynamically adjusted laser sampling frequency, v(t) is the real-time linear speed of the conveyor belt 103, and k2 is the speed adjustment coefficient. In this embodiment, k2 = 0.5.

[0077] Acquire edge contour data and images. Laser rangefinder 201 (press f) s (t) Frequency acquisition of edge contour data, industrial camera 202 captures surface images at 120fps.

[0078] In step S3, the multimodal data fusion outputs the aluminum sheet length value. Specifically, when the absolute difference between the real-time measured aluminum sheet width and the nominal aluminum sheet width is greater than 2mm, laser data compensation is performed; the defect confidence is calculated using a pre-trained ResNet-18 model; and the multi-source data is fused using a weighted Kalman filter to output the aluminum sheet length value, including the following steps:

[0079] S31, when |W act -W nom When |> 2mm, laser data compensation is performed, as shown in the following expression:

[0080] L corr =L means +k3·(W act -W nom ) 2

[0081] In the above formula, L corr The measured value of the aluminum sheet length after compensation, L means W represents the original laser measurement length value. act W is the width of the aluminum sheet measured in real time. nom Here, k is the nominal width of the aluminum sheet, and k3 is the error correction parameter. In this embodiment, k3 = 0.003.

[0082] S32, Visual defect analysis;

[0083] The defect confidence C is calculated using a pre-trained ResNet-18 model. d The output range is [0,1].

[0084] S33. Weighted Kalman filtering is used to fuse multi-source data, as shown in the following expression:

[0085]

[0086] In the above formula, L is the final aluminum sheet length value output by the fusion process. corr C is the measured value of the compensated aluminum sheet length. d For the defect confidence level, L image Here, k is the length value measured visually, and k4 is the laser data weighting coefficient. In this embodiment, k4 = 25, and the defect confidence level C... d The value ranges from 0 to 1, representing the surface defect probability output by the ResNet-18 model.

[0087] In step S4, intelligent sorting and feedback control are implemented. Specifically, the length of the aluminum sheet is calculated. and the nominal length L of the aluminum sheet nom The absolute difference, if Mark the aluminum sheet as defective, generate a sorting instruction, and record a timestamp t1.

[0088] After receiving the instruction from the industrial control computer 302, within 500ms, the non-conforming product is pushed to the temporary storage bin 402 by the pneumatic push rod 401. The actual time t2 after the sorting execution mechanism 4 completes its action is recorded, and the time delay is calculated using the following expression:

[0089] Δt = t2 - t1 - 500 ms

[0090] In the above formula, Δt is the time delay, t2 is the actual feedback time after the sorting execution mechanism 4 completes its action, and t1 is the time to generate the sorting instruction.

[0091] The emergency stop acceleration of conveyor belt 103 is dynamically adjusted based on the time delay Δt, as expressed below:

[0092]

[0093] In the above formula, a new For the adjusted emergency stop acceleration of conveyor belt 103, a old The acceleration of conveyor belt 103 before adjustment is given, and Δt is the time delay.

[0094] In step S5, data storage and model updates occur. Specifically, the detection data is stored in the self-learning database 6, including L... corr v(t), ambient temperature and ambient humidity.

[0095] The LSTM model is trained at fixed intervals, and the network structure expression is as follows:

[0096]

[0097] In the above formula, ΔL the λ is the predicted change in length due to thermal expansion. i W represents the weight coefficients from the hidden layer to the output layer. i Here, T is the weight matrix from the input layer to the hidden layer, H is the current ambient temperature, H is the current ambient humidity, t is the continuous running time of the system, and b is the weight matrix from the input layer to the hidden layer. i This is the hidden layer bias term.

[0098] The training parameters include: batch size = 64, initial learning rate = 0.001, and number of iterations = 100.

[0099] This invention can calibrate the drift of the laser rangefinder 201 and detect aluminum sheets based on their nominal length and width, improving the flexibility and accuracy of detection. It can compensate for laser data based on the aluminum sheet width and, combined with defect confidence, use weighted Kalman filtering to fuse multi-source data to output the aluminum sheet length value, improving the accuracy of aluminum sheet length detection. It can sort defective aluminum sheets based on their length value and nominal length, and can dynamically adjust the emergency stop acceleration of the conveyor belt 103 during sorting, reducing sorting delay and improving sorting efficiency. Furthermore, it can be trained based on ambient temperature and humidity to predict the length change caused by thermal expansion, reducing the false detection rate in summer production and improving detection accuracy.

[0100] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of equivalents of the invention, the invention is also intended to include these modifications and variations.

Claims

1. A high-precision automatic online detection method for the length of aluminum sheets, characterized in that, The method comprises the following steps: S1, calibrate the non-contact detection module and load parameters; S2, collect image and edge profile data, dynamically adjust laser sampling frequency based on the line speed of the conveyor belt; S3, based on laser data compensation and defect confidence, use weighted Kalman filtering to fuse multi-source data to output the length value of the aluminum sheet; S4, based on the length value of the aluminum sheet and the nominal length of the aluminum sheet, sort out the unqualified aluminum sheet, and dynamically adjust the emergency stop acceleration of the conveyor belt; In S3, when the absolute difference between the real-time measured aluminum sheet width and the aluminum sheet nominal width is greater than 2mm, laser data compensation is performed, and the expression is as follows: In the above formula, is the compensated aluminum sheet length measurement value, is the original laser measured length value, is the real-time measured aluminum sheet width, is the aluminum sheet nominal width, is the error correction parameter; The length value of the aluminum sheet is output by using weighted Kalman filtering to fuse multi-source data, and the expression is as follows: Lfinal is the final fused output aluminum sheet length value, Lfinal is the final fused output aluminum sheet length value, Lcomp is the compensated aluminum sheet length measurement value, Ldef is the defect confidence, Lvis is the visually measured length value, Llaser is the laser data weight coefficient.

2. The method according to claim 1, wherein, In S1, the laser range finder drift amount mm is executed, and the calibration expression is as follows: in the above formula, is the measured value of the calibrated laser range finder, is the measured value of the original laser range finder, is the temperature compensation coefficient, is the amount of change in the ambient temperature.

3. The method according to claim 1, wherein, In S2, the sampling frequency is dynamically adjusted based on the line speed, and the expression is as follows: in the above formula, is the dynamic adjusted laser sampling frequency, is the real-time conveying belt linear velocity, is the speed adjustment coefficient.

4. The method according to claim 1, wherein, In S4, the emergency stop acceleration of the conveyor belt is dynamically adjusted, and the expression is as follows: in the above formula, is the adjusted conveyor sudden stop acceleration, is the unadjusted conveyor sudden stop acceleration, is the time delay amount, is the actual time after the sorting actuator action is completed, is the time for generating the sorting instruction.

5. A high-precision automatic online detection system for the length of aluminum sheet, which is suitable for the high-precision automatic online detection method for the length of aluminum sheet according to any one of claims 1-4, characterized in that, It comprises: A conveying device for conveying aluminum sheets and realizing speed closed-loop control; A non-contact detection module for collecting aluminum sheet image and edge profile data; A data processing unit for calculating dynamic sampling frequency and generating sorting instructions; A sorting execution mechanism for receiving sorting instructions to sort out unqualified products; A dynamic compensation module for adjusting laser sampling frequency and edge compensation coefficient based on real-time line speed, connected to the conveying device and the data processing unit; A self-learning database for storing historical detection data and training LSTM prediction model, connected to the data processing unit through Ethernet interface.

6. The high-precision automatic online detection system for the length of aluminum sheets according to claim 5, characterized in that, The conveying device comprises a conveyor belt, a driving motor and an encoder, and the output end of the encoder is connected to the input end of the dynamic compensation module.

7. The high-precision automatic online detection system for the length of aluminum sheets according to claim 6, characterized in that, The non-contact detection module is composed of several groups of high-precision laser range finders and industrial cameras, the laser range finders are symmetrically distributed on both sides of the conveyor belt at an inclination angle of 15°, the distance between measurement points is 1.2 times the nominal width of the aluminum sheet, and the data output ends of the laser range finders and the industrial cameras are connected to the FPGA chip of the data processing unit.

8. The high-precision automatic online detection system for the length of aluminum sheets according to claim 5, characterized in that, The sorting execution mechanism comprises a pneumatic push rod and a temporary storage bin, and the pneumatic push rod is connected to the industrial computer of the data processing unit through the RS485 bus.

Images