Some parts seem to have similarities in the way they are processed, i.e., processing steps, involved streamlines, measurement duration ( e.g., Part 185485 and Part 9185. For visualization, you may refer to the post by JeffH).
\n\nSo do we need to classify parts first as certain kinds of parts are more likely to fail the quality control? Then check out the measurements?
\n\nAlso, are they assembling a single product?
", "messages": [ { "id": "135218", "postDate": "09/12/2016 07:19:32", "content": "Some parts seem to have similarities in the way they are processed, i.e., processing steps, involved streamlines, measurement duration ( e.g., Part 185485 and Part 9185. For visualization, you may refer to the post by JeffH).
\n\nSo do we need to classify parts first as certain kinds of parts are more likely to fail the quality control? Then check out the measurements?
\n\nAlso, are they assembling a single product?
", "rawMarkdown": "Some parts seem to have similarities in the way they are processed, i.e., processing steps, involved streamlines, measurement duration ( e.g., Part 185485 and Part 9185. For visualization, you may refer to the post by [JeffH][1]). \r\n\r\nSo do we need to classify parts first as certain kinds of parts are more likely to fail the quality control? Then check out the measurements?\r\n\r\nAlso, are they assembling a single product?\r\n\r\n [1]: https://www.kaggle.com/jeffhebert/bosch-production-line-performance/visualize-part-history/code", "votes": null }, { "id": "135243", "postDate": "09/12/2016 13:12:26", "content": "Hi WangshengLiu,
\n\nYour idea sounds good !
\nBoth of them have only L0 and L3, don't have L1 and L2.
Wangsheng - I wrote this script to address the question whether or not they are assembling a single product. It looks to me like no. At any given time, S30 is processing a run/batch of similarly processed parts vs. a mixture of the different "types". There could be some other explanation that I haven't considered, but this seems most likely.
\n\n", "rawMarkdown": "Wangsheng - I wrote this script to address the question whether or not they are assembling a single product. It looks to me like no. At any given time, S30 is processing a run/batch of similarly processed parts vs. a mixture of the different \"types\". There could be some other explanation that I haven't considered, but this seems most likely.\r\n\r\n[Assembly?](https://www.kaggle.com/jpmiller/bosch-production-line-performance/assembly)", "votes": null }, { "id": "135406", "postDate": "09/13/2016 21:42:51", "content": "Classifying components sounds a good idea. \nBut beware of some interesting features of the data set: it seems that a faulty component is most of the time followed by the same component (same measures) which is not faulty. So if you search for a nearest neighbour based on components characteristics you will almost always find a successful component that has the exact same profile as a faulty component... Then a first step would require to eliminate the reprocessing of faulty components first (some are even reprocessed, and fail again, and are reprocessed again). \nI'm trying another approach: finding the overall context in which the machine is likely to produce a faulty component. This means not just consider component by component, but groups of components which have a particular feature and are processed close to each other.
", "rawMarkdown": "Classifying components sounds a good idea. \r\nBut beware of some interesting features of the data set: it seems that a faulty component is most of the time followed by the same component (same measures) which is not faulty. So if you search for a nearest neighbour based on components characteristics you will almost always find a successful component that has the exact same profile as a faulty component... Then a first step would require to eliminate the reprocessing of faulty components first (some are even reprocessed, and fail again, and are reprocessed again). \r\nI'm trying another approach: finding the overall context in which the machine is likely to produce a faulty component. This means not just consider component by component, but groups of components which have a particular feature and are processed close to each other.", "votes": null } ], "comments": [ { "id": 135243, "author_name": "toshik", "author_url": "", "post_date": "09/12/2016 13:12:26", "content": "Hi WangshengLiu,
\n\nYour idea sounds good !
\nBoth of them have only L0 and L3, don't have L1 and L2.
Wangsheng - I wrote this script to address the question whether or not they are assembling a single product. It looks to me like no. At any given time, S30 is processing a run/batch of similarly processed parts vs. a mixture of the different "types". There could be some other explanation that I haven't considered, but this seems most likely.
\n\n", "votes": null, "replies": [] }, { "id": 135406, "author_name": "jencreve", "author_url": "", "post_date": "09/13/2016 21:42:51", "content": "Classifying components sounds a good idea. \nBut beware of some interesting features of the data set: it seems that a faulty component is most of the time followed by the same component (same measures) which is not faulty. So if you search for a nearest neighbour based on components characteristics you will almost always find a successful component that has the exact same profile as a faulty component... Then a first step would require to eliminate the reprocessing of faulty components first (some are even reprocessed, and fail again, and are reprocessed again). \nI'm trying another approach: finding the overall context in which the machine is likely to produce a faulty component. This means not just consider component by component, but groups of components which have a particular feature and are processed close to each other.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "135218": "Some parts seem to have similarities in the way they are processed, i.e., processing steps, involved streamlines, measurement duration ( e.g., Part 185485 and Part 9185. For visualization, you may refer to the post by [JeffH][1]). \r\n\r\nSo do we need to classify parts first as certain kinds of parts are more likely to fail the quality control? Then check out the measurements?\r\n\r\nAlso, are they assembling a single product?\r\n\r\n [1]: https://www.kaggle.com/jeffhebert/bosch-production-line-performance/visualize-part-history/code", "135243": "Hi WangshengLiu,\r\n\r\nYour idea sounds good !