{
  "id": 75397,
  "title": "Some materials and thoughts",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/75397",
  "author_name": "Rand Xie",
  "post_date": "2018-12-21T05:19:18.580000",
  "votes": 53,
  "comment_count": 0,
  "views": 0,
  "content": "<p>It is amazing that Kaggle hosts a competition on PHM (prognostics and health monitoring). In this post, I would like to share some materials and thoughts, as I had been working in this field for a few years.</p>\n\n<p><strong>Signal processing</strong></p>\n\n<p>Matlab actually provides amazing materials on signal processing. When I studied signal processing, I learned a lot by reading through their websites. The Mathworks' predictive maintenance toolbox gives a very good starting point on what kind of features can be extracted from signals in time domain, frequency domain and time-frequency domain. Please see\n<a href=\"https://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html\">https://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html</a> for more details.</p>\n\n<p><strong>Related competitions</strong></p>\n\n<p>You might be surprised that there are a lot of similarities between machine signals and biomedical signals. I would encourage you to take a look on the previous seizure detection competitions. A lot of techniques can be applied here.</p>\n\n<p>(1) <a href=\"https://www.kaggle.com/c/seizure-detection\">https://www.kaggle.com/c/seizure-detection</a></p>\n\n<p>(2) <a href=\"https://www.kaggle.com/c/seizure-prediction\">https://www.kaggle.com/c/seizure-prediction</a></p>\n\n<p>(3) <a href=\"https://www.kaggle.com/c/melbourne-university-seizure-prediction\">https://www.kaggle.com/c/melbourne-university-seizure-prediction</a></p>\n\n<p>In additional to Kaggle, PHM society (<a href=\"https://www.phmsociety.org/\">https://www.phmsociety.org/</a>) has been hosting yearly competitions since 2008. Some of them are related to signal processing. It is also a good source of information if you want to find out more techniques to try.</p>\n\n<p><strong>More than data science</strong></p>\n\n<p>Sometimes, we might need to jump out of the box and think in physics. As far as I know, it is usually more accurate to construct features based on physical system understanding. If you understand the failure mechanism, you might be able to construct some powerful features that's hard for model to discover. Try to use physics as prior!</p>",
  "messages": [
    {
      "id": 443148,
      "postDate": "2018-12-21T05:19:18.580Z",
      "content": "<p>It is amazing that Kaggle hosts a competition on PHM (prognostics and health monitoring). In this post, I would like to share some materials and thoughts, as I had been working in this field for a few years.</p>\n\n<p><strong>Signal processing</strong></p>\n\n<p>Matlab actually provides amazing materials on signal processing. When I studied signal processing, I learned a lot by reading through their websites. The Mathworks' predictive maintenance toolbox gives a very good starting point on what kind of features can be extracted from signals in time domain, frequency domain and time-frequency domain. Please see\n<a href=\"https://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html\">https://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html</a> for more details.</p>\n\n<p><strong>Related competitions</strong></p>\n\n<p>You might be surprised that there are a lot of similarities between machine signals and biomedical signals. I would encourage you to take a look on the previous seizure detection competitions. A lot of techniques can be applied here.</p>\n\n<p>(1) <a href=\"https://www.kaggle.com/c/seizure-detection\">https://www.kaggle.com/c/seizure-detection</a></p>\n\n<p>(2) <a href=\"https://www.kaggle.com/c/seizure-prediction\">https://www.kaggle.com/c/seizure-prediction</a></p>\n\n<p>(3) <a href=\"https://www.kaggle.com/c/melbourne-university-seizure-prediction\">https://www.kaggle.com/c/melbourne-university-seizure-prediction</a></p>\n\n<p>In additional to Kaggle, PHM society (<a href=\"https://www.phmsociety.org/\">https://www.phmsociety.org/</a>) has been hosting yearly competitions since 2008. Some of them are related to signal processing. It is also a good source of information if you want to find out more techniques to try.</p>\n\n<p><strong>More than data science</strong></p>\n\n<p>Sometimes, we might need to jump out of the box and think in physics. As far as I know, it is usually more accurate to construct features based on physical system understanding. If you understand the failure mechanism, you might be able to construct some powerful features that's hard for model to discover. Try to use physics as prior!</p>",
      "rawMarkdown": "It is amazing that Kaggle hosts a competition on PHM (prognostics and health monitoring). In this post, I would like to share some materials and thoughts, as I had been working in this field for a few years.\n\n**Signal processing**\n\nMatlab actually provides amazing materials on signal processing. When I studied signal processing, I learned a lot by reading through their websites. The Mathworks' predictive maintenance toolbox gives a very good starting point on what kind of features can be extracted from signals in time domain, frequency domain and time-frequency domain. Please see\nhttps://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html for more details.\n\n**Related competitions**\n\nYou might be surprised that there are a lot of similarities between machine signals and biomedical signals. I would encourage you to take a look on the previous seizure detection competitions. A lot of techniques can be applied here.\n\n(1) https://www.kaggle.com/c/seizure-detection\n\n(2) https://www.kaggle.com/c/seizure-prediction\n\n(3) https://www.kaggle.com/c/melbourne-university-seizure-prediction\n\nIn additional to Kaggle, PHM society (https://www.phmsociety.org/) has been hosting yearly competitions since 2008. Some of them are related to signal processing. It is also a good source of information if you want to find out more techniques to try.\n\n**More than data science**\n\nSometimes, we might need to jump out of the box and think in physics. As far as I know, it is usually more accurate to construct features based on physical system understanding. If you understand the failure mechanism, you might be able to construct some powerful features that's hard for model to discover. Try to use physics as prior!",
      "votes": 53
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "443148": "It is amazing that Kaggle hosts a competition on PHM (prognostics and health monitoring). In this post, I would like to share some materials and thoughts, as I had been working in this field for a few years.\n\n**Signal processing**\n\nMatlab actually provides amazing materials on signal processing. When I studied signal processing, I learned a lot by reading through their websites. The Mathworks' predictive maintenance toolbox gives a very good starting point on what kind of features can be extracted from signals in time domain, frequency domain and time-frequency domain. Please see\nhttps://www.mathworks.com/help/predmaint/ug/signal-based-condition-indicators.html for more details.\n\n**Related competitions**\n\nYou might be surprised that there are a lot of similarities between machine signals and biomedical signals. I would encourage you to take a look on the previous seizure detection competitions. A lot of techniques can be applied here.\n\n(1) https://www.kaggle.com/c/seizure-detection\n\n(2) https://www.kaggle.com/c/seizure-prediction\n\n(3) https://www.kaggle.com/c/melbourne-university-seizure-prediction\n\nIn additional to Kaggle, PHM society (https://www.phmsociety.org/) has been hosting yearly competitions since 2008. Some of them are related to signal processing. It is also a good source of information if you want to find out more techniques to try.\n\n**More than data science**\n\nSometimes, we might need to jump out of the box and think in physics. As far as I know, it is usually more accurate to construct features based on physical system understanding. If you understand the failure mechanism, you might be able to construct some powerful features that's hard for model to discover. Try to use physics as prior!"
  }
}