<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ganesh Krishnan on Medium]]></title>
        <description><![CDATA[Stories by Ganesh Krishnan on Medium]]></description>
        <link>https://medium.com/@gkrishnan7?source=rss-ac090ff2f7e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1dzkfZ02Rk4SOSKizR7muQ.jpeg</url>
            <title>Stories by Ganesh Krishnan on Medium</title>
            <link>https://medium.com/@gkrishnan7?source=rss-ac090ff2f7e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 12:00:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gkrishnan7/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Practical Applications of NLP in the Oil & Gas Industry]]></title>
            <link>https://medium.com/@gkrishnan7/practical-applications-of-nlp-in-the-oil-gas-industry-3b07c3a1d361?source=rss-ac090ff2f7e------2</link>
            <guid isPermaLink="false">https://medium.com/p/3b07c3a1d361</guid>
            <category><![CDATA[oil-and-gas-industry]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[oil-and-gas]]></category>
            <category><![CDATA[drilling]]></category>
            <dc:creator><![CDATA[Ganesh Krishnan]]></dc:creator>
            <pubDate>Thu, 04 Aug 2022 14:40:21 GMT</pubDate>
            <atom:updated>2022-08-04T14:40:21.822Z</atom:updated>
            <content:encoded><![CDATA[<p>When the Oil &amp; Gas industry commences drilling a well, a lot of nomenclature specific to the Oil &amp; Gas industry is used to communicate the day-to-day drilling operations. Oil &amp; Gas operators such as Chevron has made wide use of NLP techniques to absorb information from free text, and thus giving a fresh perspective in well and reservoir planning.</p><p>Risk identification and assessment are critical during the drilling stages. A lot of engineers use their past experiences to guide their approach on a certain project, and this can possibly obstruct them from taking a radically different approach. As a result, Drilling Engineers may overcompensate to make sure that event is mitigated (i.e. stuck pipe, having to fish for a logging tool, etc.).</p><p>There are two data sets that the folks in Chevron use: the Well Operations and Risk Assessment database. Drilling Engineers create daily log entries that describe the days event and categorizes them. The Well Operations database is free text description that fall into two categories: type and type detail. Well Operations database is by and large labeled data. The type and type detail provide classifications for each event that transpired during the particular day the well was drilled (<em>Natural-Language-Processing Techniques for Oil and Gas Drilling Data</em>, 2017).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1eeKd5_VTuaKTefGNZMydA.jpeg" /><figcaption>Figure 1 (Daily Completion Workover Schematic, 2000)</figcaption></figure><p>Risk Assessment data usually contains short phrases with technical jargon. Risk Assessment is unlabeled data. Steps were taken to validate the Risk Assessment data, a random set of 700 samples were hand labeled from the Risk Assessment data set. These samples were referred to as the Risk Assessment Gold data set (<em>Natural-Language-Processing Techniques for Oil and Gas Drilling Data</em>, 2017).</p><p>An NLP application was created to receive free-text risk assessment from the Drilling Engineer and show events transpired from nearby wells that have been drilled in the past. The application goes through the labeled Well Operations data to identify how many events were identified and inputted as risk assessments. This allows the powers that be to make informed decisions about what actions to take on the well that they are currently drilling.</p><p>The Risk Assessment and Well Operations Data sets are merged to compare historical data and determine the accuracy of a potential well failure based on previous information captured from particular wells and well groups surrounding the current well.</p><p>Well Operations data set is split 60% training and split evenly between development and test sets. Risk Assessment data is treated like an unlabeled test set. Preprocessing functions are applied to the free text descriptions. Statistically insignificant features were removed from the Well Operations data set if they occur less than five times. Text is made lowercase, n-gram features are extracted, and numbers and punctuations are removed. The features that are left are converted into a sparse-feature vector (<em>Natural-Language-Processing Techniques for Oil and Gas Drilling Data</em>, 2017).</p><p>To address the different styles of writing and vocabulary used between the two datasets, a set of labeled data was extracted and ran some queries to match the type and type detail to the associated Risk Assessment data. A classification model was run on Well Operations &amp; Risk Assessment Auto data.</p><p>The classifier model was run with the Well Operations and Risk Assessment Gold data set. When the class weights were recalibrated and training data was supplemented with combined data of Well Operations &amp; Risk Assessment Auto data, we saw major improvement in the results (<em>Natural-Language-Processing Techniques for Oil and Gas Drilling Data</em>, 2017).</p><p>In conclusion, the NLP application helps Drilling Engineers predict risks to their wells by looking at historical risk assessments, and help encounter unforeseen problems.</p><h3>References:</h3><p><em>Daily Completion Workover Schematic</em>. (2000). Peloton. Retrieved July 29, 2022, from <a href="https://www.peloton.com/assets/images/content/23_DailyCompletionandWorkoverSchematic_0.jpg">https://www.peloton.com/assets/images/content/23_DailyCompletionandWorkoverSchematic_0.jpg</a></p><p><em>Natural-Language-Processing Techniques for Oil and Gas Drilling Data</em>. (2017, September 30). JPT. Retrieved July 29, 2022, from <a href="https://jpt.spe.org/natural-language-processing-techniques-oil-and-gas-drilling-data">https://jpt.spe.org/natural-language-processing-techniques-oil-and-gas-drilling-data</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3b07c3a1d361" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Practical Applications of Recurrent Neural Network in Oil & Gas Industry]]></title>
            <link>https://medium.com/@gkrishnan7/practical-applications-of-recurrent-neural-network-in-oil-gas-industry-439c8eaed984?source=rss-ac090ff2f7e------2</link>
            <guid isPermaLink="false">https://medium.com/p/439c8eaed984</guid>
            <category><![CDATA[predictive-maintenance]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[oil-and-gas]]></category>
            <category><![CDATA[recurrent-neural-network]]></category>
            <category><![CDATA[oil-and-gas-industry]]></category>
            <dc:creator><![CDATA[Ganesh Krishnan]]></dc:creator>
            <pubDate>Tue, 26 Jul 2022 13:01:45 GMT</pubDate>
            <atom:updated>2022-07-26T13:01:45.522Z</atom:updated>
            <content:encoded><![CDATA[<p>Oil &amp; Gas facilities try to do preventative maintenance so that they mitigate the chance of significant equipment failure. Just like you take your car to get a periodic oil change, the same applies to oil &amp; gas equipment in the oil fields. If you don’t do periodic oil changes, you stand the risk of significant issues that could cost you a lot more money down the line. Keep in mind that preventative maintenance though done with the best of intentions can be inefficient and labor intensive.</p><p>This article summarizes the findings from a research paper titled, “Predictive Maintenance of Oil and Gas Equipment using Recurrent Neural Network” by Tayaba Abbasi, King Hann Lim, and Ke San Yam done in 2019.</p><p>The paper is making a significant pivot from preventative maintenance to predictive maintenance. Predictive maintenance is vital in that it improves equipment efficiency, and downtime reduction. An RNN-LSTM was proposed to do predictive maintenance on an air booster compressor (ABC) motor.</p><p>A two-year of historical data was evaluated of ABC motor having one minute interval. The data was obtained from a real plant. ABC motor faced significant downtime three times in a two-year time span. There is data prior to each failure. The data was divided into two categories: healthy and unhealthy data. The unhealthy data was subdivided into three types of faults: shutdown, outliers, and out of range fault. Shutdown fault refers to a zero reading from the motor. Outliers are when you have an abnormal reading from the motor. The out of range fault refers to data outside the range of the motor reading. The unhealthy data was filtered and put into upper and lower range of each parameter. Fourteen parameters were collected from the ABC motor. The healthy data got normalized using a Z score formula (Abbasi et al., 2019).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/459/1*QD1PZwlIouXWrVT0TWKLrA.png" /><figcaption><em>Figure 1 (Abbasi et al., 2019)</em></figcaption></figure><p>5700 experimental data points were used to run in the RNN-LSTM model. The data is divided into training, validation, and testing to understand the accuracy of the model. Sixty percent of the data was devoted to training, and the rest was split evenly between validation and testing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/468/1*7XUJBLIr5wxWei6zeIHu4A.png" /><figcaption><em>Figure 2 (Abbasi et al., 2019)</em></figcaption></figure><p>Two training algorithms were used to compare the simulation results for the RNN-LSTM model. Lavenberg-Marquardt proves to be a faster function that outputs a lower RMSE (root mean square error) value. Lavenberg-Marquardt are trained on functions with sum squared error types and thus making it faster to train neural networks measure on that kind of error (Abbasi et al., 2019).</p><p>Table 3 shows that the model was trained with different epochs and number of neurons in hidden layer. As illustrated in Table 3, there was a point of diminishing returns if you increased the number of neurons and the epochs. If the model exceeded 15 neurons and 750 epochs, then your model will overfit and increase the RMSE. If the model underperforms in the testing data, then overfitting exists (Abbasi et al., 2019).</p><p>In conclusion, predictive maintenance can be used to forecast equipment failure especially in the Oil &amp; Gas industry. Fourteen parameters were obtained from the ABC motor and subsequentially fed into an RNN-LSTM model. The model was used to predict the signal for 30 minutes sampling. A huge benefit to using LSTM is that it doesn’t require expert knowledge and feature engineering. LSTM can capture meaningful results and dependencies.</p><h3>References:</h3><p>Abbasi, T., Lim, K. H., &amp; Yam, K. S. (2019). Predictive Maintenance of Oil and Gas Equipment using Recurrent Neural Network. <em>IOP Conference Series: Materials Science and Engineering</em>, <em>495</em>, 012067. <a href="https://doi.org/10.1088/1757-899x/495/1/012067">https://doi.org/10.1088/1757-899x/495/1/012067</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=439c8eaed984" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Practical Applications of Convolutional Neural Networks in Oil & Gas Industry]]></title>
            <link>https://medium.com/@gkrishnan7/practical-applications-of-convolutional-neural-networks-in-oil-gas-industry-e6b30a2e6f8?source=rss-ac090ff2f7e------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6b30a2e6f8</guid>
            <category><![CDATA[oil-and-gas]]></category>
            <category><![CDATA[oil-and-gas-industry]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[multi-phase-flow]]></category>
            <category><![CDATA[convolution-neural-net]]></category>
            <dc:creator><![CDATA[Ganesh Krishnan]]></dc:creator>
            <pubDate>Sun, 24 Jul 2022 15:43:15 GMT</pubDate>
            <atom:updated>2022-07-24T15:43:15.600Z</atom:updated>
            <content:encoded><![CDATA[<p>Good day Medium Family! My name is Ganesh Krishnan. This is my first Medium post. I am a Petroleum Engineer by trade. Currently getting my Masters in Applied Artificial Intelligence at the University of San Diego. My future articles will be centered around the following themes: AI, Energy, and Crypto. So stay tuned folks!</p><p>I wanted to start writing and get myself out there. This article should be treated for educational purposes only. Today I will be talking about an interesting paper that I read titled “Measure of Gas-Oil Two Phase Flow Patterns by Using CNN Algorithm Based on Dual ECT Sensors with Venturi Tube” by Zhuoqun Xu, Fan Wu, Xinmeng Yang, and Yi Li.</p><p>The Oil &amp; Gas industry does a lot of extensive research into multi-phase flow particularly to improve the safety, reliability, efficiency, and reduction of maintenance. This is of utmost importance in terms of an engineering and economical point of view. Accurate measurement of oil-gas two phase parameters is a barrier to production of hydrocarbons.</p><p>This case study used a dual-ECT (Electric Capacitance tomography) to collect the capacitance values before and after the venturi tube. Electric capacitance tomography consists of three components: sensors, the system of data acquisition, and the system of computer imaging. It gives the opportunity to visualize the flow configuration. The data acquisition system applies voltage to each electrode plate of the sensor, and then gets the capacitance values between any two plates. The capacitance value is then inputted into the computer imaging (Xu et al., 2020).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*RgZX8jKP-DNKm03OLs8GwA.jpeg" /><figcaption>Figure 1 (2022)</figcaption></figure><p>As seen on the figure below, the current common flow patterns in a pipeline include bubble flow, stratified flow, wavy flow, slug flow, and annular flow. An image reconstruction algorithm is used to capture flow patterns in different working environments. The input for the CNN model is to capture the flow patterns before and after the venturi tube, the output is the predicted oil and gas flow rates.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/380/1*_XnEUmWJ7ws1qIDuLrbhuA.jpeg" /><figcaption>Figure 2 (Two Phase Flow in a Horizontal Pipe, 2017)</figcaption></figure><p>A Convolutional Neural Network (CNN) is proposed to predict the oil &amp; gas flow rate. The authors used an ElasticNet regression as a loss function to compensate for overfitting problems. The authors used an Inception V3 model of the CNN algorithm to solve the non-linear mapping of oil-gas two-phase flow parameters and changes in the flow patterns. Inception V3 structure and branches are very complicated. It is very suitable for solving high dimensional and non-linear relationship in flow patterns and corresponding GVF (gas-void fraction). The input is image pixel data, so the dimensionality of the data is high and that is also a large dataset to analyze. The Inception V3 network integrates two-dimensional data into smaller convolutions. The flow patterns in before and after the venturi tube were trained to predict oil and gas flow rates, and GVF (Xu et al., 2020).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/334/1*tnuguiylyuH6m-DqDKqkUg.png" /><figcaption>Figure 3 (Xu et al., 2020)</figcaption></figure><p>The authors did a comparison between SVM (Support Vector Machine) and CNN algorithm when they ran the experiment. In the figure below with the list of tables, CNN algorithm provides the most accurate prediction of oil and gas flow rates. CNN network is suitable for regression of high dimensional image data. The average relative error for an oil flow rate before and after the venturi tube is 4.6%, the predicted relative error of gas flow rate is 1.43%, and the average relative error for GVF is 1.6% (Xu et al., 2020).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/420/1*WNLwG-u5w4MClG9d-hZ1HA.png" /><figcaption>Figure 4 (Xu et al., 2020)</figcaption></figure><p>In conclusion, CNN algorithm improves upon the accuracy of oil and gas two-phase flow. The only caveat to mass adoption within the Oil &amp; Gas industry is the computational power required to run the CNN algorithm. Much work needs to be done to improve computational efficiency.</p><h3>REFERENCES:</h3><p><em>Two Phase Flow in a Horizontal Pipe</em>. (2017). ResearchGate. Retrieved July 15, 2023, from <a href="https://www.researchgate.net/figure/REPRESENTATIVE-TWO-PHASE-FLOW-PATTERNS-IN-A-HORIZONTAL-PIPE_fig1_322395593">https://www.researchgate.net/figure/REPRESENTATIVE-TWO-PHASE-FLOW-PATTERNS-IN-A-HORIZONTAL-PIPE_fig1_322395593</a></p><p>U., U., Lengkapku, L. P., &amp; Google+, J. M. O. (2022, July 15). <em>Electrical Capacitance Volume Tomography (ECVT) part #1</em>. Blogspot. <a href="http://etekno.blogspot.com/2012/07/electrical-capacitance-volume.html">http://etekno.blogspot.com/2012/07/electrical-capacitance-volume.html</a></p><p>Xu, Z., Wu, F., Yang, X., &amp; Li, Y. (2020). Measurement of Gas-Oil Two-Phase Flow Patterns by Using CNN Algorithm Based on Dual ECT Sensors with Venturi Tube. <em>Sensors</em>, <em>20</em>(4), 1200. <a href="https://doi.org/10.3390/s20041200">https://doi.org/10.3390/s20041200</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6b30a2e6f8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>