###### tags: `transformer-pipeline` `pin-inspector` `introduction` # pin-inspector.introction **整脚后引脚检测** 用于检测整脚后的变压器是否仍存在引脚不良 [TOC] ## temp about k: 电压的一个单位对应物理的一，一丝=0.01mm，`丝/1个读数单位` about x offset: 以内侧激光测距仪为基准，外测的相对偏移量（光栅尺方向、电机方向） ``` k:-6.32496744e-04 ``` 可能会有这种情况，圆会拟合到上半部分，考虑用判定一下是否圆心小于平均值 ![](https://i.imgur.com/4cMcixu.png) **实验结果缓存** 整个激光测距仪的数据非常的依赖与针脚的距离是否合适，调整时保证变压器在正中位置时能够稳定即可 ## codes ### [pin-analyzer.test_vt3](http://gitlab.sho.hypereal.com/robotic/pin-analyzer/tree/test_vt3) ```c //record data test_client.py //for calibration pin_inspector_get_k.py pin_inspector_get_x_offset.py pin_inspector_get_y_offset.py ``` ### [transformer-pipeline.test_pin_inspector_online](http://gitlab.sho.hypereal.com/robotic/transformer-pipeline/tree/pin_inspector_test_machine) ## calibration > [origin version](https://docs.google.com/document/d/1EyHpowRUUNk0ByNS8flumy6wC0IqT6TcH2gvWeCUfpk/edit#) [name=陈易平] ### calibrate k value step 1, in pin_analyzer, retrieve realtime adc value by using ``` python3 pin_analyzer/test_client.py ``` step 2, write down the ADC value and the total thickness of sliders, put them to a text file step 3, redo to get more ADC value using different totalthickness of sliders, then we get the all record into a file, something like ``` 2935.3446896459764 15 2938.887459163944 15 6134.095006333756 13 6134.096939795986 13 9268.84698979932 11 9268.829188612573 11 12431.636109073937 9 12432.950996733116 9 ... // 5..10 ``` step 4, calibrate with these record using: ``` python3 pin_inspector_get_k.py ``` step 5, check k, about 6 * 10^-4 ### calibrate x offset step 1, place a cylinder on the pin inspector platform step 2, make sure that this cylinder can be detected by lasers lie both sides step 3, record data ``` cd build mkdir sample /bin/sample_pin_inspector ``` step 4, check data, channels & circles, ensure that you can write down a range which only contains the two circles, asume that rangea is $[8000, 15000]$，and chanels are $5$ and $7$ ``` python3 tools/visualize_pin_inspector.py ``` ![](https://i.imgur.com/DcjmNbS.png) step 5, write config ``` lower_limit uppder_limit chanels //from outter side to the inner side ``` step 6, calibrate & sample output ``` python3 pin_inspector_get_x_offset.py 32.47104626852157 //in mm ``` ![](https://i.imgur.com/mMe9kfj.png) ### calibrate y offset config step1, use the phneumatic claw to seize a standard slider, then ``` cd build mkdir sample /bin/sample_pin_inspector ``` step 2, check data, channels & 2 target lines, ensure that you can write down a range which only contains the two lines ``` python3 tools/visualize_pin_inspector.py ``` chanels are $5$ and $7$, rangea is $[10000, 15500]$ ![](https://i.imgur.com/8kzXMjw.png) step 3, write config ``` kThicknessOfMiddleSlider kThicknessOfSideSlider lower_limit uppder_limit chanels //from outter side to the inner side ``` step 4, calibrate ``` python3 pin_inspector_get_y_offsert.py -13.59174925040592 //mm ``` ## test ( python version ) the number in following picture shows 10 pins of A451 ``` 6_7 __ _ _8_9 inner side ########### ####A451### ########### 0_1_2_3_4_5 outter side ``` step 1, do "二次元" calibration，get the position and picture of pins ``` 圆1 内容 测量值 名义值 超差值 上公差 下公差 状态 圆心X 38.8593 38.8593 0.0000 圆心Y 45.8104 45.8104 0.0000 .... ... ``` ![](https://i.imgur.com/WaMRWCG.jpg) step 2, use sliders to adjust the Y-position of laser devices, make sure that pins of both sides of the transformer can be detected by laser devices step 3, record data ``` sample_pin_inspector.cpp ``` step 4, test :::warning there are more than one entrance inside the main function of script, make sure that you are runing the script using the entrance you wanted ::: ```python # python3 pin_visualizer.py if __name__ == '__main__': show_extract_round_pins(folder, kSpeed, kObjId, kPosId, kFrontAdcId, kBackAdcId, True) # show_extract_flat_pins(folder, kSpeed, kObjId, kPosId, kFrontAdcId, kBackAdcId, True) # show(folder, kSpeed, kObjId) ``` ![](https://i.imgur.com/hS9ReX2.png) step 5, compare ( python version ) 将二次元的数据和上一个步骤的数据放在一起（最好记录成实验文档，备查 ![](https://i.imgur.com/dbKr45O.png) 拷贝出来到一个文件里准备运行对比程序 ![](https://i.imgur.com/NAHRAr9.png) 配置好后运行运行代码： ``` python3 minimize_compare.py ``` 效果较差： ![](https://i.imgur.com/rciEQ9m.png) 效果较好： ![](https://i.imgur.com/9e3fQE8.png) ## compare ( cpp version ) ## 项目主要问题 激光测距仪+算法拟合出的针脚足够精确： - 多次实验存在的误差？（之前初步测试过标准差约在 0.几 丝） - 和二次元结果的误差？ - 可容忍的误差范围？ - 激光测距仪（高速模式）在要求速度下的稳定性？ 算法可优化？ - 比如之前计算 loss 时选择的标准（avg 或 max） - 优化范围过滤，过滤掉无关区间 如何保证 **硬件问题** **软件问题**