![\"Faults\"](\"https://i.imgur.com/0Rq6BTX.png\")
Hi, I think it would help either FE or network design and preprocessing to know exactly what we are looking for.
\n\nHere's some signals labelled with faults, and I've highlighted the areas where I THINK the fault is:
\n\n
and here more NON FAULTS, some of them have \"interesting\" areas which I've marked in green:
\n\n![\"Not](\"https://i.imgur.com/MVkeSqL.png\")
Any PD experts can throw some light here?
", "messages": [ { "id": 455865, "postDate": "2019-01-14T17:59:46.430Z", "content": "Hi, I think it would help either FE or network design and preprocessing to know exactly what we are looking for.
\n\nHere's some signals labelled with faults, and I've highlighted the areas where I THINK the fault is:
\n\n
and here more NON FAULTS, some of them have \"interesting\" areas which I've marked in green:
\n\n
Any PD experts can throw some light here?
", "rawMarkdown": "Hi, I think it would help either FE or network design and preprocessing to know exactly what we are looking for.\n\nHere's some signals labelled with faults, and I've highlighted the areas where I THINK the fault is:\n\n![Faults][1]\n\nand here more NON FAULTS, some of them have \"interesting\" areas which I've marked in green:\n\n![Not faults][2]\n\nAny PD experts can throw some light here?\n\n [1]: https://i.imgur.com/0Rq6BTX.png\n [2]: https://i.imgur.com/MVkeSqL.png", "votes": 48 }, { "id": 463209, "postDate": "2019-01-29T16:11:57.090Z", "content": "@tvantuch any word on this? I believe it is in the interest of your organization who is sponsoring the competition to help competitors achieve best results by throwing as much info as you can about the problem.
\n\nIf labeling was automatic and invisible (even confusing) to the human eye, I'd understand. This happened in the IEEE's Signal Processing Society Camera Model Identification, no human can easily explain labels but there's a real ground truth captured deterministically with no error.
\n\nIn this case it looks (please correct if not the case), labels are annotated by trained person. I understand there must be annotation guidelines. If they are not available, can you at least comment on why some triplets of signals are labeled as faulty where only one of them looks as such?
", "rawMarkdown": "@tvantuch any word on this? I believe it is in the interest of your organization who is sponsoring the competition to help competitors achieve best results by throwing as much info as you can about the problem.\n\nIf labeling was automatic and invisible (even confusing) to the human eye, I'd understand. This happened in the IEEE's Signal Processing Society Camera Model Identification, no human can easily explain labels but there's a real ground truth captured deterministically with no error.\n\nIn this case it looks (please correct if not the case), labels are annotated by trained person. I understand there must be annotation guidelines. If they are not available, can you at least comment on why some triplets of signals are labeled as faulty where only one of them looks as such? ", "votes": 10 }, { "id": 456142, "postDate": "2019-01-15T08:22:10.547Z", "content": "Some additional questions and observations.
\n\nTake these signals, which correspond to the contiguous 3 phase:\n![\"cont\"](\"http://i.imgur.com/5I27RWt.png\")
I have a hard time reconciling it with this from this paper :\n![\"PD](\"http://i.imgur.com/7WssgoJ.png\")
which is also mentioned here:\n![\"sig](\"http://i.imgur.com/qpmifH7.png\")
why? All 3 phases above are labelled as faulty. In the signals above the only visible patterns that happen in all 3 phases are the big highlighted peaks, and they happen at the same time.
\n\nBy definition all 3 phases are shifted in phase, and we've seen in some EDAs that 3-phase faults (which in my initial exploration I attribute to synced PD patterns in all 3 phases...) are significantly more common than single-phase faults; so this could go against the observation that the faults happen more often at specific parts of the waveform (if this would be true, the PD patterns will happen at different times in 3-phase faults).
\n\nI am obviously missing something obvious here...
", "rawMarkdown": "Some additional questions and observations.\n\nTake these signals, which correspond to the contiguous 3 phase:\n![cont][1]\n\nI have a hard time reconciling it with this from [this paper](https://www.dropbox.com/s/2ltuvpw1b1ms2uu/A%20Complex%20Classification%20Approach%20of%20Partial%20Discharges%20from%20Covered%20Conductors%20in%20Real%20Environment%20%28preprint%29.pdf?dl=0) :\n![PD patter location][2]\n\nwhich is also mentioned [here](http://dspace.vsb.cz/bitstream/handle/10084/133114/VAN431_FEI_P1807_1801V001_2018.pdf):\n![sig zones][3]\n\nwhy? All 3 phases above are labelled as faulty. In the signals above the only visible patterns that happen in all 3 phases are the big highlighted peaks, and they happen at the same time. \n\nBy definition all 3 phases are shifted in phase, and we've seen in some EDAs that 3-phase faults (which in my initial exploration I attribute to synced PD patterns in all 3 phases...) are significantly more common than single-phase faults; so this could go against the observation that the faults happen more often at specific parts of the waveform (if this would be true, the PD patterns will happen at different times in 3-phase faults).\n\nI am obviously missing something obvious here...\n\n [1]: http://i.imgur.com/5I27RWt.png\n [2]: http://i.imgur.com/7WssgoJ.png\n [3]: http://i.imgur.com/qpmifH7.png", "votes": 5 }, { "id": 457210, "postDate": "2019-01-17T04:44:13.610Z", "content": "I wish I could push this post to forum top :P
\n\nI also stared at id_measurment 76, which is labeled faulty at all 3 phases, for a very long time.\nThe large pulse appearing at all 3 phases is common in negative samples.\nYet the small hairy peaks are appearing when the sine wave voltage is moving back to zero, which also contradicts what the literature says...
", "rawMarkdown": "I wish I could push this post to forum top :P\n\nI also stared at id_measurment 76, which is labeled faulty at all 3 phases, for a very long time.\nThe large pulse appearing at all 3 phases is common in negative samples.\nYet the small hairy peaks are appearing when the sine wave voltage is moving back to zero, which also contradicts what the literature says...", "votes": 6 }, { "id": 462344, "postDate": "2019-01-28T06:40:01.073Z", "content": "My guess is that if any phase is faulty, all three phases are labeled as faulty. Otherwise the triplet 228, 229 and 230 do not make sense since only 230 seems to look like it is suffering from partial discharge. In power systems, single phase faults are much more common than three phase faults. If my guess is correct, then i agree, its very confusing.
", "rawMarkdown": "My guess is that if any phase is faulty, all three phases are labeled as faulty. Otherwise the triplet 228, 229 and 230 do not make sense since only 230 seems to look like it is suffering from partial discharge. In power systems, single phase faults are much more common than three phase faults. If my guess is correct, then i agree, its very confusing.", "votes": 1, "replies": [ { "id": 463333, "postDate": "2019-01-29T20:28:32.883Z", "content": "The data speaks otherwise.
", "rawMarkdown": "The data speaks otherwise.", "votes": 1 } ] }, { "id": 455966, "postDate": "2019-01-14T22:01:34.080Z", "content": "More, these ones have fault (area highlighted is my hypothesis of fault):\n![\"fault\"](\"http://i.imgur.com/lmkjJJy.png\")
\nand these one doesn't, but to me it looks like it does:\n![\"no](\"http://i.imgur.com/W04v450.png\")
I just found the same thing and enter to discussions to post it. \nParticularly signals 228,229,230. All of them has target=1 (e.g. PD) and they are from same line (id_measurement=76), but I see 228 and 229 very suspicious, it did not like PD, it is more RPI. Especially 228 is very 'clean'.
", "rawMarkdown": "I just found the same thing and enter to discussions to post it. \nParticularly signals 228,229,230. All of them has target=1 (e.g. PD) and they are from same line (id_measurement=76), but I see 228 and 229 very suspicious, it did not like PD, it is more RPI. Especially 228 is very 'clean'.", "votes": 2 }, { "id": 456022, "postDate": "2019-01-15T01:12:46.213Z", "content": "Below discussion might be helpful for your analysis of signals. \nhttps://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/75771
\n\nIn his paper, there are some descriptions about PD pattern, DSI and RPI.\nAttached figure is from the above discussion, and it shows the patterns of several signals.\nSome regions of signals with several peaks (pulses), which you highlighted, may not contain exact PD patterns. They could be some interferences like DSI, RPI, and in that cases, they may not have target value 1.
", "rawMarkdown": "Below discussion might be helpful for your analysis of signals. \nhttps://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/75771\n\nIn his paper, there are some descriptions about PD pattern, DSI and RPI.\nAttached figure is from the above discussion, and it shows the patterns of several signals.\nSome regions of signals with several peaks (pulses), which you highlighted, may not contain exact PD patterns. They could be some interferences like DSI, RPI, and in that cases, they may not have target value 1.", "votes": 2 }, { "id": 458124, "postDate": "2019-01-18T20:38:43.973Z", "rawMarkdown": "", "isDeleted": true }, { "id": 480554, "postDate": "2019-02-28T10:32:30.073Z", "content": "Thanks, Andres, Really Helpful
", "rawMarkdown": "Thanks, Andres, Really Helpful" } ], "comments": [ { "id": 463209, "author_name": "Andrés Miguel Torrubia Sáez", "author_url": "", "post_date": "2019-01-29T16:11:57.090000", "content": "@tvantuch any word on this? I believe it is in the interest of your organization who is sponsoring the competition to help competitors achieve best results by throwing as much info as you can about the problem.
\n\nIf labeling was automatic and invisible (even confusing) to the human eye, I'd understand. This happened in the IEEE's Signal Processing Society Camera Model Identification, no human can easily explain labels but there's a real ground truth captured deterministically with no error.
\n\nIn this case it looks (please correct if not the case), labels are annotated by trained person. I understand there must be annotation guidelines. If they are not available, can you at least comment on why some triplets of signals are labeled as faulty where only one of them looks as such?
", "votes": 10, "replies": [] }, { "id": 456142, "author_name": "Andrés Miguel Torrubia Sáez", "author_url": "", "post_date": "2019-01-15T08:22:10.547000", "content": "Some additional questions and observations.
\n\nTake these signals, which correspond to the contiguous 3 phase:\n
I have a hard time reconciling it with this from this paper :\n
which is also mentioned here:\n
why? All 3 phases above are labelled as faulty. In the signals above the only visible patterns that happen in all 3 phases are the big highlighted peaks, and they happen at the same time.
\n\nBy definition all 3 phases are shifted in phase, and we've seen in some EDAs that 3-phase faults (which in my initial exploration I attribute to synced PD patterns in all 3 phases...) are significantly more common than single-phase faults; so this could go against the observation that the faults happen more often at specific parts of the waveform (if this would be true, the PD patterns will happen at different times in 3-phase faults).
\n\nI am obviously missing something obvious here...
", "votes": 5, "replies": [] }, { "id": 457210, "author_name": "YLChan", "author_url": "", "post_date": "2019-01-17T04:44:13.610000", "content": "I wish I could push this post to forum top :P
\n\nI also stared at id_measurment 76, which is labeled faulty at all 3 phases, for a very long time.\nThe large pulse appearing at all 3 phases is common in negative samples.\nYet the small hairy peaks are appearing when the sine wave voltage is moving back to zero, which also contradicts what the literature says...
", "votes": 6, "replies": [] }, { "id": 462344, "author_name": "Bikiran Guha", "author_url": "", "post_date": "2019-01-28T06:40:01.073000", "content": "My guess is that if any phase is faulty, all three phases are labeled as faulty. Otherwise the triplet 228, 229 and 230 do not make sense since only 230 seems to look like it is suffering from partial discharge. In power systems, single phase faults are much more common than three phase faults. If my guess is correct, then i agree, its very confusing.
", "votes": 1, "replies": [ { "id": 463333, "author_name": "عثمان", "author_url": "", "post_date": "2019-01-29T20:28:32.883000", "content": "The data speaks otherwise.
", "votes": 1, "replies": [] } ] }, { "id": 455966, "author_name": "Andrés Miguel Torrubia Sáez", "author_url": "", "post_date": "2019-01-14T22:01:34.080000", "content": "More, these ones have fault (area highlighted is my hypothesis of fault):\n\nand these one doesn't, but to me it looks like it does:\n
I just found the same thing and enter to discussions to post it. \nParticularly signals 228,229,230. All of them has target=1 (e.g. PD) and they are from same line (id_measurement=76), but I see 228 and 229 very suspicious, it did not like PD, it is more RPI. Especially 228 is very 'clean'.
", "votes": 2, "replies": [] }, { "id": 456022, "author_name": "Kim", "author_url": "", "post_date": "2019-01-15T01:12:46.213000", "content": "Below discussion might be helpful for your analysis of signals. \nhttps://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/75771
\n\nIn his paper, there are some descriptions about PD pattern, DSI and RPI.\nAttached figure is from the above discussion, and it shows the patterns of several signals.\nSome regions of signals with several peaks (pulses), which you highlighted, may not contain exact PD patterns. They could be some interferences like DSI, RPI, and in that cases, they may not have target value 1.
", "votes": 2, "replies": [] }, { "id": 458124, "author_name": "", "author_url": "", "post_date": "2019-01-18T20:38:43.973000", "content": "", "votes": 0, "replies": [] }, { "id": 480554, "author_name": "JohnJayChou", "author_url": "", "post_date": "2019-02-28T10:32:30.073000", "content": "Thanks, Andres, Really Helpful
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "455865": "Hi, I think it would help either FE or network design and preprocessing to know exactly what we are looking for.\n\nHere's some signals labelled with faults, and I've highlighted the areas where I THINK the fault is:\n\n![Faults][1]\n\nand here more NON FAULTS, some of them have \"interesting\" areas which I've marked in green:\n\n![Not faults][2]\n\nAny PD experts can throw some light here?\n\n [1]: https://i.imgur.com/0Rq6BTX.png\n [2]: https://i.imgur.com/MVkeSqL.png", "463209": "@tvantuch any word on this? I believe it is in the interest of your organization who is sponsoring the competition to help competitors achieve best results by throwing as much info as you can about the problem.\n\nIf labeling was automatic and invisible (even confusing) to the human eye, I'd understand. This happened in the IEEE's Signal Processing Society Camera Model Identification, no human can easily explain labels but there's a real ground truth captured deterministically with no error.\n\nIn this case it looks (please correct if not the case), labels are annotated by trained person. I understand there must be annotation guidelines. If they are not available, can you at least comment on why some triplets of signals are labeled as faulty where only one of them looks as such? ", "456142": "Some additional questions and observations.\n\nTake these signals, which correspond to the contiguous 3 phase:\n![cont][1]\n\nI have a hard time reconciling it with this from [this paper](https://www.dropbox.com/s/2ltuvpw1b1ms2uu/A%20Complex%20Classification%20Approach%20of%20Partial%20Discharges%20from%20Covered%20Conductors%20in%20Real%20Environment%20%28preprint%29.pdf?dl=0) :\n![PD patter location][2]\n\nwhich is also mentioned [here](http://dspace.vsb.cz/bitstream/handle/10084/133114/VAN431_FEI_P1807_1801V001_2018.pdf):\n![sig zones][3]\n\nwhy? All 3 phases above are labelled as faulty. In the signals above the only visible patterns that happen in all 3 phases are the big highlighted peaks, and they happen at the same time. \n\nBy definition all 3 phases are shifted in phase, and we've seen in some EDAs that 3-phase faults (which in my initial exploration I attribute to synced PD patterns in all 3 phases...) are significantly more common than single-phase faults; so this could go against the observation that the faults happen more often at specific parts of the waveform (if this would be true, the PD patterns will happen at different times in 3-phase faults).\n\nI am obviously missing something obvious here...\n\n [1]: http://i.imgur.com/5I27RWt.png\n [2]: http://i.imgur.com/7WssgoJ.png\n [3]: http://i.imgur.com/qpmifH7.png", "457210": "I wish I could push this post to forum top :P\n\nI also stared at id_measurment 76, which is labeled faulty at all 3 phases, for a very long time.\nThe large pulse appearing at all 3 phases is common in negative samples.\nYet the small hairy peaks are appearing when the sine wave voltage is moving back to zero, which also contradicts what the literature says...", "462344": "My guess is that if any phase is faulty, all three phases are labeled as faulty. Otherwise the triplet 228, 229 and 230 do not make sense since only 230 seems to look like it is suffering from partial discharge. In power systems, single phase faults are much more common than three phase faults. If my guess is correct, then i agree, its very confusing.", "455966": "More, these ones have fault (area highlighted is my hypothesis of fault):\n![fault][1]\nand these one doesn't, but to me it looks like it does:\n![no fault][2]\n\n [1]: http://i.imgur.com/lmkjJJy.png\n [2]: http://i.imgur.com/W04v450.png", "456432": "I just found the same thing and enter to discussions to post it. \nParticularly signals 228,229,230. All of them has target=1 (e.g. PD) and they are from same line (id_measurement=76), but I see 228 and 229 very suspicious, it did not like PD, it is more RPI. Especially 228 is very 'clean'.", "456022": "Below discussion might be helpful for your analysis of signals. \nhttps://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/75771\n\nIn his paper, there are some descriptions about PD pattern, DSI and RPI.\nAttached figure is from the above discussion, and it shows the patterns of several signals.\nSome regions of signals with several peaks (pulses), which you highlighted, may not contain exact PD patterns. They could be some interferences like DSI, RPI, and in that cases, they may not have target value 1.", "458124": "", "480554": "Thanks, Andres, Really Helpful" } }