Part testing device

Abstract

This invention discloses a component testing apparatus for testing at least one component under test. The apparatus includes a real-time controller, a servo driver, a first electric cylinder module, and a second electric cylinder module. The real-time controller uses artificial intelligence to provide a real-time control signal containing at least one optimal proportional-integral-derivative (PID) control parameter. The servo driver is coupled to the real-time controller and configured to generate a first drive signal and a second drive signal based on the real-time control signal. The first and second electric cylinder modules are coupled to the servo driver and are driven by the first and second drive signals, respectively, to apply a first force and a second force to the component under test according to the at least one optimal PID control parameter.

Description

Technical Field

[0001] This invention relates to component testing, and in particular to a component testing device that uses artificial intelligence (AI) to achieve precise control of the sinusoidal output force of an electric cylinder module. Background Technology

[0002] like Figure 1 As shown, the traditional component testing device 1 is designed to perform component testing simply by controlling the solenoid valve switch with a programmable logic controller (PLC).

[0003] Taking fatigue testing of bicycle parts as an example, the testing equipment needs to apply a stable sinusoidal force with a frequency below 10Hz to the part under test, and the error between the peak and trough values ​​must be at least within ±5% of the test force (as required by ISO standards). Sometimes, even a more stringent ±2% error range is targeted. Since different types of parts under test have different rigidities, the test control parameters required for testing different types of parts must be adjusted accordingly. However, currently, experienced automatic control engineers still manually adjust the test control parameters based on the actual test conditions.

[0004] However, if the rigidity of the part under test is found to be too hard or too soft during actual testing, the originally set range of test control parameters will not be applicable and will inevitably need to be readjusted. However, the technicians who perform fatigue testing on the part under test generally do not have this expertise and cannot adjust the appropriate test control parameters in real time, which leads to the risk of traditional part testing equipment going out of control.

[0005] Therefore, the aforementioned problems encountered by existing technologies still need to be solved. Summary of the Invention

[0006] In view of this, the present invention proposes a component testing device to effectively solve the above-mentioned problems encountered by the prior art.

[0007] A preferred embodiment of the present invention provides a component testing apparatus. In this embodiment, the component testing apparatus is used to test at least one component to be tested. The component testing apparatus includes a real-time controller, a servo driver, a first electric cylinder module, and a second electric cylinder module. The real-time controller is configured to provide a real-time control signal containing at least one optimal proportional-integral-derivative (PID) control parameter using artificial intelligence (AI). The servo driver is coupled to the real-time controller and configured to generate a first drive signal and a second drive signal based on the real-time control signal. The first electric cylinder module is coupled to the servo driver and configured to be driven by the first drive signal to apply a first force to the component to be tested according to the at least one optimal PID control parameter. The second electric cylinder module is coupled to the servo driver and configured to be driven by the second drive signal to apply a second force to the component to be tested according to the at least one optimal PID control parameter.

[0008] In one embodiment, the first force applied by the first electric cylinder module is to stretch the part to be tested, and the second force applied by the second electric cylinder module is to stretch the part to be tested.

[0009] In one embodiment, the first force applied by the first electric cylinder module is to compress the part to be tested, and the second force applied by the second electric cylinder module is to compress the part to be tested.

[0010] In one embodiment, the first force applied by the first electric cylinder module is to stretch the part under test and the second force applied by the second electric cylinder module is to compress the part under test, or the first force applied by the first electric cylinder module is to compress the part under test and the second force applied by the second electric cylinder module is to stretch the part under test.

[0011] In one embodiment, the first force applied by the first electric cylinder module and the second force applied by the second electric cylinder module are respectively half-wave inverse actions performed on the part under test.

[0012] In one embodiment, the first force applied by the first electric cylinder module and the second force applied by the second electric cylinder module are respectively applied to the part under test with a fixed slope to perform a static pressure holding test.

[0013] In one embodiment, the at least one most suitable PID control parameter is generated by a multi-layer perceptron (MLP) neural network.

[0014] In one embodiment, the multilayer perceptron neural network uses collected PID parameter combinations and corresponding rigidity data samples of the test part for supervised learning and verifies its prediction results, and then generates the corresponding at least one most suitable PID control parameter according to the type and rigidity of the test part.

[0015] In one embodiment, the first electric cylinder module and the second electric cylinder module apply a first force and a second force to the part under test as sinusoidal forces, and the accuracy of the first force and the second force is judged by the peak value of the sinusoidal force, the number of cycles required to reach the stable control force, and the total harmonic distortion (THD).

[0016] Compared to existing technologies, the component testing device proposed in this invention can be applied to any testing situation requiring dynamic force application via PID control. It can be used to test any component such as bicycle frames, handlebars, seatposts, cranks, and forks that require force waveforms controlled by PID, such as half-sine wave frame pedal fatigue testing and sine wave handlebar fatigue testing. The component testing device proposed in this invention can be trained and learned through artificial intelligence (AI) to provide the most suitable PID control parameters in real time according to different types of test components or different rigidities, thereby effectively solving the problem that traditionally, professionals need to manually adjust the PID control parameters to achieve optimal performance. Thus, AI is used to achieve precise control of the sine wave output of the electric cylinder module. Attached Figure Description

[0017] Figure 1 This diagram illustrates a traditional component testing device designed to perform component testing simply by using a programmable logic controller (PLC) to control solenoid valve switches.

[0018] Figure 2 A functional block diagram of a part testing apparatus according to a preferred embodiment of the present invention is shown.

[0019] Figure 3 This diagram illustrates the sinusoidal force applied to the part to be tested by the part testing apparatus of the present invention.

[0020] Figure 4 A schematic diagram of a part testing apparatus according to another preferred embodiment of the present invention is shown.

[0021] Figure 5 This diagram illustrates how an electric cylinder module applies force to the part under test through a load cell to perform fatigue testing.

[0022] Figure 6 The diagram shows a bicycle handlebar being tested by the part testing device.

[0023] Figure 7 An example of a training data form is shown, obtained from testing bicycle handlebars.

[0024] Figure 8 An example of a training data form for an AI model. Detailed Implementation

[0025] A preferred embodiment of the present invention is a component testing device. In this embodiment, the component testing device of the present invention can learn through artificial intelligence (AI) to provide the most suitable PID control parameters in real time according to different types of components under test or different rigidities. Therefore, it can be applied to any test situation that requires dynamic force application by PID control, such as any test of bicycle frames, handlebars, seatposts, cranks, forks, etc., that requires force waveforms controlled by PID, including half-sine wave frame pedal fatigue test and sine wave handlebar fatigue test, etc., thereby effectively solving the problem that traditionally, professionals need to manually adjust the PID control parameters to achieve optimal performance, thus realizing fully automated AI real-time component testing.

[0026] Please refer to Figure 2 , Figure 2 A functional block diagram of the part testing device 2 in this embodiment is shown. (See diagram below.) Figure 2 As shown, the component testing apparatus 2 is used to test at least one component CP to be tested. The component testing apparatus 2 includes a real-time controller 20, a servo driver 22, a first electric cylinder module 24, and a second electric cylinder module 26. The real-time controller 20 is coupled to the servo driver 22, the first electric cylinder module 24, and the second electric cylinder module 26. The servo driver 22 is coupled to the real-time controller 20, the first electric cylinder module 24, and the second electric cylinder module 26. The first electric cylinder module 24 is coupled to both the real-time controller 20 and the servo driver 22. The second electric cylinder module 26 is coupled to both the real-time controller 20 and the servo driver 22.

[0027] Real-time controller 20 is configured to provide real-time control signal SC to servo driver 22. Servo driver 22 is configured to generate a first drive signal SD1 and a second drive signal SD2 based on real-time control signal SC. First electric cylinder module 24 is configured to apply a first force F1 to the part CP under test based on the first drive signal SD1. Second electric cylinder module 26 is configured to apply a second force F2 to the part CP under test based on the second drive signal SD2. When real-time controller 20 receives a first feedback signal FB1 from first electric cylinder module 24 and a second feedback signal FB2 from second electric cylinder module 26, real-time controller 20 dynamically adjusts real-time control signal SC based on the first feedback signal FB1 and the second feedback signal FB2.

[0028] In one embodiment, both the first force F1 applied by the first electric cylinder module 24 to the part CP to be tested and the second force F2 applied by the second electric cylinder module 26 to the part CP to be tested can be... Figure 3 The accuracy of the sinusoidal force shown, and whether the first force F1 applied by the first electric cylinder module 24 to the part CP under test and the second force F2 applied by the second electric cylinder module 26 to the part CP under test are accurate, can be judged by the peak value of their sinusoidal force, the number of cycles required to reach the stable control force, and the total harmonic distortion (THD). For example, if the error between the peak value and the trough value of the sinusoidal wave of the first force F1 and the second force F2 is less than ±2% of the test force, then the first force F1 and the second force F2 applied by the first electric cylinder module 24 and the second electric cylinder module 26 are judged to be accurate, but this is not a limitation.

[0029] It should be noted that since the first electric cylinder module 24 and the second electric cylinder module 26 can operate independently, they can also test two different parts under test, but this is not a limitation.

[0030] In another embodiment, the real-time controller 20 may apply artificial intelligence (AI) to provide a real-time control signal SC containing at least one proportional-integral-derivative (PID) control parameter best suited to the part under test (CP). When the servo driver 22 receives the real-time control signal SC, the servo driver 22 drives the first electric cylinder module 24 and the second electric cylinder module 26 accordingly based on the PID control parameter best suited to the part under test (CP) in the real-time control signal SC. The first force F1 and the second force F2, respectively, are applied to the part under test (CP) according to the at least one PID control parameter best suited to the part under test (CP) to best match the stiffness of the part under test (CP), so as to perform a precise fatigue test on the part under test (CP).

[0031] In detail, PID is an automatic control method widely used in various fields to achieve precise control. In this invention, when the real-time controller 20 adopts the PID control method, the proportional algorithm (P) in the PID controller has a large weight when the peak value of the sine wave, the number of cycles required to reach stable control force, and the total harmonic distortion (THD) are still far from the target value. After proportional calculation (P), because the distance is still large, the value of P will be large, causing the algorithm to adjust the waveform of the sine wave and move towards the target value at a relatively fast speed until it gradually slows down when it is close to the target value. As it gets closer to the target value, the integral algorithm (I) in the PID controller gradually gains a larger weight. Since the integral is the sum of the errors with the target value over a period of time, the integral value will naturally be smaller as it gets closer to the target value, allowing the algorithm to adjust the moving speed accordingly. Finally, when the difference from the target value becomes smaller, the rate of change between the expected value and the actual value is calculated, which is the derivative algorithm (D) in PID. When the rate of change is less than the predetermined value, the algorithm determines that the waveform of the sine wave has reached the precise waveform of the target. Therefore, the P value, I value and D value at this time are the final control values.

[0032] It should be noted that the real-time controller 20 of the part testing device 2 of the present invention can provide the most suitable PID control parameters for the part CP to be tested in real time according to different types of parts CP to be tested or their different rigidities through AI learning and training.

[0033] For example, the real-time controller 20 can first perform supervised learning using a multi-layer perceptron (MLP) neural network with the collected PID control parameter combinations and corresponding rigidity data samples of the part under test (CP), and then verify whether its prediction results are correct. If its prediction results are verified to be correct, the real-time controller 20 can generate the most suitable PID control parameters for the CP under test according to the type and rigidity of the CP under test, and then transmit a real-time control signal SC containing the most suitable PID control parameters to the servo driver 22. The servo driver 22 can then drive the first electric cylinder module 24 and the second electric cylinder module 26 to apply the first force F1 and the second force F2, respectively, which best match the rigidity of the CP under test, to perform a precise fatigue test on the CP under test.

[0034] In detail, a Multilayer Perceptron (MLP) neural network is a type of feedforward neural network that contains at least three layers (input layer, hidden layer, and output layer). It utilizes a "backpropagation" technique to achieve supervised learning. In an MLP neural network, each layer can contain many independent neurons. There are no connections between neurons within the same layer, but there are corresponding connections between neurons in the layers above and below each other. In other words, each neuron in a lower layer learns a weight value from each neuron in an upper layer to represent the connection strength between the neurons in those layers. For classification tasks, this connection strength may be a unique feature belonging to a particular category.

[0035] When the MLP neural network is applied to the part testing device of the present invention, the tester must identify the characteristics of the PID test parameter combination and use these characteristics to determine whether the PID test parameter combination is suitable for the rigidity of the part under test. If the MLP neural network is trained using the collected data samples corresponding to the rigidity of the part under test and the corresponding data samples of the PID test parameter combination, and its prediction results are verified to be correct, the part testing device of the present invention can then provide the most suitable PID test control parameters according to different parts under test or different rigidities using artificial intelligence.

[0036] Please refer to Figure 4 In another embodiment, the component testing device 4 includes a real-time controller 40, a first servo driver 41, a second servo driver 42, a first servo motor 43, a first electric cylinder module 44, a second servo motor 45, a second electric cylinder module 46, a first load unit 47, a second load unit 48, a first bridge amplifier 49, and a second bridge amplifier 50.

[0037] Real-time controller 40 is coupled to first servo driver 41, second servo driver 42, first bridge amplifier 49, and second bridge amplifier 50. First servo driver 41 is coupled to real-time controller 40 and first servo motor 43. Second servo driver 42 is coupled to real-time controller 40 and second servo motor 45. First servo motor 43 is coupled to first servo driver 41 and first electric cylinder module 44. First electric cylinder module 44 is coupled to first servo motor 43 and first load cell 47. Second servo motor 45 is coupled to second servo driver 42 and second electric cylinder module 46. Second electric cylinder module 46 is coupled to second servo motor 45 and second load cell 48. First load cell 47 is coupled to first electric cylinder module 44, first bridge amplifier 49, and the part under test (DUT) CP. Second load cell 48 is coupled to second electric cylinder module 46, second bridge amplifier 50, and the DUT CP. First bridge amplifier 49 is coupled to first load cell 47 and real-time controller 40. The second bridge amplifier 50 is coupled to the second load unit 48 and the real-time controller 40, respectively.

[0038] The real-time controller 40 can receive AI control parameters PM transmitted by the remote control device RM via Ethernet EN, but is not limited thereto. The real-time controller 40 can provide real-time control signals SC to the first servo driver 41 and the second servo driver 42 according to the AI ​​control parameters PM. The first servo driver 41 and the second servo driver 42 generate first drive signals SD1 and second drive signals SD2 respectively to the first servo motor 43 and the second servo motor 45 according to the real-time control signals SC. The first servo motor 43 and the second servo motor 45 are driven by the first drive signals SD1 and the second drive signals SD2 respectively, thereby activating the first electric cylinder module 44 and the second electric cylinder module 46 to apply a first force F1 and a second force F2 to the test part CP, respectively, through the first load unit 47 and the second load unit 48, to the test part CP, with the stiffness best matching that of the test part CP. The first feedback signal FB1 and the second feedback signal FB2 related to the first force F1 and the second force F2 of the first load unit 47 and the second load unit 48 are transmitted to the real-time controller 40 through the first bridge amplifier 49 and the second bridge amplifier 50. The real-time controller 40 dynamically adjusts the real-time control signal SC based on the first feedback signal FB1 and the second feedback signal FB2.

[0039] It should be noted that, as Figure 5 As shown, since the first electric cylinder module 44 and the second electric cylinder module 46 operate independently, the first electric cylinder module 44 and the second electric cylinder module 46 can also test two different parts under test (the first part under test CP1 and the second part under test CP2) simultaneously through the first load unit 47 and the second load unit 48, but this is not a limitation.

[0040] Please refer to Figure 6 When the part to be tested, CP, is a bicycle handlebar, the part testing device of the present invention applies a first force F1 and a second force F2 to the first region A1 and the second region A2 of the part to be tested, CP, respectively, such as the left and right regions of a bicycle handlebar, but is not limited thereto. In practical applications, the part testing device of the present invention may include, but is not limited to, the following various test modes: (1) The first force F1 is to stretch the first area A1 of the part to be tested CP and the second force F2 is to stretch the second area A2 of the part to be tested CP. (2) The first force F1 is to compress the first area A1 of the part to be tested CP and the second force F2 is to compress the second area A2 of the part to be tested CP; (3) The first force F1 is to stretch the first region A1 of the part to be tested CP and the second force F2 is to compress the second region A2 of the part to be tested CP; (4) The first force F1 is to compress the first area A1 of the part to be tested CP and the second force F2 is to stretch the second area A2 of the part to be tested CP. (5) The first force F1 and the second force F2 are respectively applied to the first region A1 and the second region A2 of the part under test CP, performing half-wave inversion; and (6) The first force F1 and the second force F2 are forces with a constant slope applied to the first region A1 and the second region A2 of the part to be tested CP, respectively, to perform static pressure holding test.

[0041] In one embodiment, the upper layer of the software can be written in LabVIEW, and the data processing, training, and prediction results of an AI model written in Python can be called. Taking the handlebar fatigue test specified in section 4.9 of ISO 4210-5 for bicycles as an example, it stipulates that dynamic forces of the same and opposite phases are applied to both ends of the handlebar. Trained engineers manually adjust the test control parameters for various bicycle parts (steels, cranks, forks) to ensure that the peak value of the sine wave controlled by the electric cylinder module, the number of cycles required to reach stable control force, and the total harmonic distortion (THD) meet the requirements. This allows for the acquisition of ideal test control parameters for various bicycle parts, which are then input into the AI ​​model for learning and training. The accuracy of the sine wave force provided by the electric cylinder module is indicated by the peak value of the sine wave, the number of cycles required to reach stable control force, and the total harmonic distortion (THD). The peak value of the sine wave needs to remain stable over a long period and within ±1% of the test force error, while a smaller THD indicates a waveform closer to a perfect sine wave. The peak value and THD of the sinusoidal force provided by the electric cylinder module can be calculated using the LabVIEW analysis module to determine whether the control of the electric cylinder module is ideal. When the user actually uses the part testing device of this invention, a tension-compression action needs to be performed after the part to be tested is installed to determine its rigidity. Then, after selecting the restraint conditions, AI calculation is performed to obtain the AI ​​control parameter PM. The real-time controller 40 can receive the AI ​​control parameter PM through the Ethernet network EN, thereby realizing high-speed and high-precision waveform control of the sinusoidal force.

[0042] For example, if the part to be tested is a bicycle handlebar, and the force is applied at a distance of 50mm / 100mm / 150mm from both ends of the handlebar, the test force is a sinusoidal force with a frequency of 3Hz and an amplitude of 280N, using the following combination of P and I values ​​(adjust P=0.01, 0.03, 0.05, adjust I=0.01, 0.03, 0.05). Apply a static force of 100N to the part to be tested, read the displacement (mm), and calculate the stiffness (N / mm). Record the peak error, valley error, THD, and number of cycles after reaching stable control. Finally, obtain the data sheet used for training as shown in the figure. Figure 7 As shown.

[0043] Please refer to Figure 8 , Figure 8 An example of a data form for training an AI model. For example... Figure 8 As shown, the data used for AI model training can include, but is not limited to, stiffness values, P-values, I-values, D-values, maximum (Max), minimum (Min), THD, and number of cycles. Once the model is built, it can calculate the maximum (Max), minimum (Min), THD, and number of cycles corresponding to the stiffness of the PID controller based on the range of its input stiffness values ​​and P and I values. The optimal P and I parameters are selected by filtering THD values. As indicated by the dashed box, the AI ​​model's judgment rule is that the minimum of the squared average of the maximum (Max), minimum (Min), and THD is the optimal solution. Using the P and I values ​​obtained from the AI ​​model, it can be observed that the number of cycles after reaching stable control is significantly reduced.

[0044] Compared to existing technologies, the component testing device proposed in this invention can be applied to any testing situation requiring dynamic force application via PID control. It can be used to test any component such as bicycle frames, handlebars, seatposts, cranks, and forks that requires force waveforms controlled by PID, such as half-sine wave frame pedal fatigue testing and sine wave handlebar fatigue testing. The component testing device proposed in this invention can be trained and learned through AI to provide the most suitable PID control parameters in real time according to different types of test components or different rigidities, thereby effectively solving the problem that traditionally, professionals need to manually adjust the PID control parameters to achieve optimal performance. This allows AI to be used to achieve precise control of the sine wave output of the electric cylinder module.

[0045] [Symbol Explanation] 1…Traditional component testing equipment PLC (Programmable Logic Controller) 2… Parts testing device 20… Real-time controller 22…Servo Driver 24…First Electric Cylinder Module 26…Second Electric Cylinder Module CP… Parts to be tested SC… Real-time control signal SD1…First drive signal SD2…Second drive signal F1…First Force F2…Second Power FB1…First Feedback Signal FB2…Second Feedback Signal 4… Parts testing device 40… Real-time controller 41…First Servo Driver 42…Second servo drive 43…First servo motor 44…First Electric Cylinder Module 45…Second servo motor 46…Second Electric Cylinder Module 47…First Load Unit 48…Second Load Unit 49…First Bridge Amplifier 50…Second Bridge Amplifier PM…AI control parameters EN…Ethernet RM…Remote Control Device CP1…First part to be tested CP2…Second part to be tested A1…First Area A2…Second Zone

Claims

1. A component testing apparatus for testing at least one component to be tested, characterized in that, The part testing device includes: A real-time controller is configured to apply artificial intelligence (AI) to provide a real-time control signal containing at least one optimal proportional-integral-derivative (PID) control parameter; A servo driver, coupled to the real-time controller, is configured to generate a first drive signal and a second drive signal based on the real-time control signal. A first electric cylinder module, coupled to the servo driver, is configured to be driven by the first drive signal to apply a first force to the part under test according to the at least one most suitable PID control parameter; as well as A second electric cylinder module, coupled to the servo drive, is configured to be driven by the second drive signal to apply a second force to the part under test according to the at least one most suitable PID control parameter.

2. The part testing device as described in claim 1, characterized in that, The first force applied by the first electric cylinder module is to stretch the part under test, and the second force applied by the second electric cylinder module is to stretch the part under test.

3. The part testing device as described in claim 1, characterized in that, The first force applied by the first electric cylinder module is to compress the part under test, and the second force applied by the second electric cylinder module is to compress the part under test.

4. The part testing device as described in claim 1, characterized in that, The first force applied by the first electric cylinder module is to stretch the part under test and the second force applied by the second electric cylinder module is to compress the part under test, or the first force applied by the first electric cylinder module is to compress the part under test and the second force applied by the second electric cylinder module is to stretch the part under test.

5. The part testing apparatus as described in claim 1, characterized in that, The first force applied by the first electric cylinder module and the second force applied by the second electric cylinder module are respectively half-wave inverse actions performed on the part under test.

6. The part testing apparatus as described in claim 1, characterized in that, The first force applied by the first electric cylinder module and the second force applied by the second electric cylinder module are respectively applied to the part under test with a constant slope to perform a static pressure holding test.

7. The part testing apparatus as described in claim 1, characterized in that, The most suitable PID control parameter is generated by a multi-layer perceptron (MLP) neural network.

8. The part testing apparatus as described in claim 7, characterized in that, The multilayer perceptron neural network uses the collected PID parameter combinations and corresponding rigidity data samples of the test part to perform supervised learning and verify its prediction results. Then, it generates the corresponding at least one most suitable PID control parameter according to the type and rigidity of the test part.

9. The part testing apparatus as described in claim 1, characterized in that, The first electric cylinder module and the second electric cylinder module apply a first force and a second force to the part under test, which are sinusoidal forces. The accuracy of the first force and the second force is judged by the peak value of the sinusoidal force, the number of cycles required to reach the stable control force, and the total harmonic distortion (THD).

Images