and H.R.S. In an approach to detect faults early, LS-SVM Regression is applied to a Vertical Form Fill and Seal. As individual algorithms present their own set of strengths and weaknesses the reader is recommended to study the individual algorithm as well. Wilson's path in the pump industry has spanned nearly 50 years. Confusion Matrix helps us to visualize model performance. Traini E., Bruno G., DAntonio G., Lombardi F. Machine Learning Framework for Predictive Maintenance in Milling. Zhou D., Al-Durra A., Zhang K., Ravey A., Gao F. Online remaining useful lifetime prediction of proton exchange membrane fuel cells using a novel robust methodology. 2325 May 2012; pp. An adaptive-order particle filter for remaining useful life prediction of aviation piston pumps. The WFE supports this and gives great results. Stochastic algorithms are mainly used within degradation models and the most common Bayesian network algorithms are Particle Filters, Kalman Filters, and hidden Markov models [29,30]. In this way, the Bayesian method can present the current state of the system, but can also evaluate future trends before a given threshold. We will discuss only the best model in detail. An overview of the process for developing a successful machine learning model, be it in a PdM setting or another. Zhang et al. These initial steps are essential when partnering with a technology provider and can help companies develop and adopt a predictive maintenance solution for their pumps that is robust and accurate. Clustering represents a category of unsupervised algorithms, which objective is to find clusters within a data set. An alternative approach proposed by Zhou et al. The hold-out method is often recommended for larger data sets. MathWorks is pleased to present an industry-tested, hands-on workshop on Predictive Maintenance using Machine Learning. The contractor wanted to reduce unplanned downtime and unexpected failures. Displacing liquid has always been a need for humans to easily transport e.g., water and it is a key part of the modern infrastructure. This is possible with a reduced computational complexity from applying a compressed RNN. He may be reached at contact@predictronics.com. [69]. [56], LS-SVM is applied in an alternative manner to a distillation process. Determine what data has been collected, if any, and review what data could be available from the pumps controller. Each of the 13 types of faults could then further be divided into more specific reasons for the fault happening. The amount of models available has rapidly increased and gaining an overview can be a challenge, thus finding an appropriate model requires testing. Ali J.B., Chebel-Morello B., Saidi L., Malinowski S., Fnaiech F. Accurate bearing remaining useful life prediction based on Weibull distribution and artificial neural network. Qiu et al. Merkt et al. Though, the order of the steps might be defined or prioritised differently, the same steps tend to be included. Model Evaluation considers the performance of the model on an unseen test data set. No free lunch theorems for optimization. [48] delves into this issue. This follows the increase in deployment of ICT enabling data and knowledge sharing. This is common within e.g., batteries [30,33,54]. The machinery in consideration is that of an ion-implanter tool. Failure types by category for a centrifugal pump. Kim G., Kim H., Zio E., Heo G. Application of particle filtering for prognostics with measurement uncertainty in nuclear power plants. Is mathematics important to the computer profession and Machine Learning? [70], applies fuzzy C-means alongside ANFIS to detect the RUL of a distillation column. Lughofer and Sayed-Mouchaweh [24] presents the concepts and applications of the various PdM approaches and introduces a set of case studies. Continuous improvements in information and communication technology, ICT, have increased data accessibility and in turn created a basis for Industry 4.0. A deep dive into applications found on pumping systems will be provided, before elaborating on how this is or could be applied to a CHP setting. [66] to try GAF and Moving Average Mapping, respectively, and converts the data into two-dimensions. Zhang B., Zhang S., Li W. Bearing performance degradation assessment using long short-term memory recurrent network. Hundi P., Shahsavari R. Comparative studies among machine learning models for performance estimation and health monitoring of thermal power plants. Furthermore, industrial data sets, typically, do not have a labelling metric, hence it requires the developer to dive into expert knowledge or company ERP systems to identify faulty data, which in turn can introduce bias. If unlabelled, the typical problem type will be that of unsupervised, which allows for clustering. David Siegel is the chief technology officer at Predictronics. Zhou K.B., Zhang J.Y., Shan Y., Ge M.F., Ge Z.Y., Cao G.N. [45], a waveform entropy can be calculated: Here WFE is the waveform entropy, Wt is the waveform factor at time instance t, and M is the length of the sliding window. We will discuss only the best model in detail, To get the best performance from machine learning models we will tune hyperparameters using Cross Validation technique. Another advantage being that it can achieve multiple outputs. The paper touches upon the issues of data gathering and storage, as it is an energy intensive process. Smart Things: Ubiquitous Computing User Experience Design. The below GIF demonstrates the working of the application. A comprehensive predictive maintenance management program uses the most cost-effective tools (e.g., vibration monitoring, thermography, tribology) to obtain the actual operating condition of critical plant systems and based on this actual data schedules all maintenance activities on an as-needed basis. Therefore, PdM strives to identify trends, anomalies, degradation at an early stage, so that sufficient counter measurements can be deployed. In case sensor_01 also we can see most of the values missing in Broken state. A predictive maintenance system for integral type faults based on support vector machines: An application to ion implantation; Proceedings of the 2013 IEEE International Conference on Automation Science and Engineering (CASE); Madison, WI, USA. [44], considers algorithms across the categories of statistical and ANN. This is seen in cars going x kilometers before going for service or jet engines doing x cycles. This is especially related to domain-specific areas, as general literature does not seem to be lacking. Can come from running frequencies of the pump, Thrust directed towards the center of the, Thrust imposed on the shaft in either an inboard or, An unavoidable fault is the recirculation of some, Opening of the lapped faces results in solids, Stems from unbalanced moving parts, particles of, A typical indication of the pumping system, Clogging of piping system or the impeller, Summation Wavelet Extreme Learning Machine, Long-Short Term Memory Recurrent Neural Network, Simplified Fuzzy Adaptive Resonance Theory Map (Neural Network), Least-squares Support-vector Machine Nonlinear Autoregressive, Backward Smoothing Square Root Cubature Kalman Filter. The findings on pumping systems can be extended to other fields. First, understanding what type of data you have at your disposition will allow for narrowing down what type of model will be suitable and what information you can achieve. A common algorithm is ARIMA, but it is also possible to apply regression techniques [22]. The team of analytics experts was able to pull crucial features from the data by considering vibration patterns in the frequency and time-frequency domain. Prior art conducting reviews within the field of PHM can be found in Table 1. 195200. The authors have tried to take this grouping-by-similarity approach, but the authors are aware that some might fall into two categories or not fall into any category. 3 0 obj Avoiding this is desirable. The issue then becomes to estimate the RUL from the available data. The following definition of PdM is provided by Mobley [18]; Predictive Maintenance is a philosophy or attitude that, simply stated, uses the actual operating condition of plant equipment and systems to optimize total plant operation. Table 4 presents an overview of detectable errors with their corresponding data tags that might be of interest. To avoid these unexpected failures, many companies increase preventative maintenance and create aggressive inspection schedules. We reserve the right to substitute course instructors as necessary. The field dealing with these complex systems and benefits from the development in technology is the study of Prognostics and Health Management, PHM. Section 3 will then delve into the applications found within pumping systems and CHPs, this is done by presenting an initial overview of common faults within pumps. The paper initially presented a thorough introduction to what PdM offers and gave a framework on how to prepare and develop PdM models. Therefore time based 5-fold cross validation is implemented and the training data is split as shown below, In the kth split, first k folds are used as training data and (k+1)th fold is used as test data. It is a Binary Classification problem. This paper was written in connection with a project on Predictive Maintenance within thermal power plants. Webster J.A., McNay D.A., Lundy D., De Sapio V., De Sapio V., Fan Q., Fravel D., Houck G., Lenchitz H., Mapes C., et al. An overview of various methods that can be chosen depending on what information is desired and to what extent data is available. The second phase, the order of an ARMA model is estimated and deployed to filter the linear degradation elements in the data. Machine learning models operate with different parameters to allow for flexibility and to increase or reduce bias. On the Use of Predictive Models for Improving the Quality of Industrial Maintenance: An Analytical Literature Review of Maintenance Strategies; Proceedings of the 2019 Federated Conference on Computer Science and Information Systems (FedCSIS); Leipzig, Germany. Approach for a Holistic Predictive Maintenance Strategy by Incorporating a Digital Twin. This will in many cases be proven to be wrong with many systems being dynamic. A set of challenges for predictive maintenance were then presented before outlining future trends. [68], k-means is deployed in a supervised setting to see the performance of detecting faults on a Selective Laser Melting machine tool. An article released by Collins and Davis on PowerEngineering [8], gives a great overview of things to consider before investing in one type of pump and states that a power plant of 300 MW will on average have approximately 100 pumps installed. Where Ahmed et al. Liu Q., Dong M., Lv W., Geng X., Li Y. Failure of these pumps not only results in unexpected operation delays and increased costs, but it can lead to dangerous oil and gas leaks, impacting labor safety and the environment. A novel Switching Unscented Kalman Filter method for remaining useful life prediction of rolling bearing. Hu Q., Ohata E.F., Silva F.H., Ramalho G.L., Han T., Filho P.P.R. This allows the developer to determine how the model potentially will perform prior to being deployed. This is challenged by the No free lunch theorem, as a component at one plant might differ slightly from another and then, in turn, prove that another model might be more suitable. This is considered out of scope for this paper. The downside is that asset operators do not know whether the maintenance is deployed efficiently [2,4]. The base case is the current PvM scheme. The models then helped determine key indicators of pump seal failure, as well as establish the accuracy and necessity of the sensors. Isaksson A.J., Harjunkoski I., Sand G. The impact of digitalization on the future of control and operations. The k-fold method splits the data such that a kth of the data is used to predict upon and the rest for training, this being iterated k times. [61], SVM is utilised to classify whether the machinery is in need of maintenance. 1618 November 2016; pp. Zhang S., Zhai B., Guo X., Wang K., Peng N., Zhang X. Synchronous estimation of state of health and remaining useful lifetime for lithium-ion battery using the incremental capacity and artificial neural networks. CNN is applied and finds an accuracy of 95% and 98% for the two fans. Another challenge lies in the task of determining an accurate RUL. endobj The paper by Langone et al. The literature presented on pumps is not limited to centrifugal pumps as they play a significant role in various systems. The filter deployed is called a Discrete Bayes Filter. Rather than giving a single estimated output on the current system health, it gives a probability distribution of possible likely options [22]. Copyright Cahaba Media Group, Inc. All Rights Reserved. The reactive maintenance approach allows components and machinery to run till failure. This approach moreover benefited from the lowered energy consumption. How to overcome challenges and tap into proactive problem-solving in oil and gas applications. The web application is hosted on AWS EC-2 instance and can be accessed using the below link. Thus, the purpose and novelty of this paper is the focus it provides on specific topic. These services will be increasingly common. The paper finds the model to predict RUL with high confidence. A more detailed overview can be seen in Figure 6. With the deployment of communication technology, sensors have been a key source of data, sending and storing data in databases. In a paper by Tse et al. This will continuously increase the possibilities to solve increasingly complex tasks and open up for various new architectures and algorithms from the ML field. Algorithms typically rely on a centroid or hierarchical approach to determine the clustering of data that reflects the shortest distance internally and the largest distance between clusters. The proposed method is not reliant on run-to-failure data and can predict the RUL even with one channel failing to provide data. Combined heat and power plants, CHP, are an important asset in current energy infrastructures, as they can provide power, grid stability, and heating simultaneously. Machine learning methods tend to be grouped by their way of functioning, e.g., tree-based models, such as decision trees, or neural networks with their various architectures. sharing sensitive information, make sure youre on a federal Carvalho T.P., Soares F.A., Vita R., da Francisco P.R., Basto J.P., Alcal S.G.S. Mechanical engineers, Maintenance engineers, Asset reliability engineers, Predictive maintenance teams, Data scientists engaged in predictive maintenance, The workshop attendees will need a laptop and an internet connection. Relative to the amount of time it takes to establish a digital twin, a data-driven model can be developed fast. This is done by splitting the degradation into 5 categories. Gerum P.C.L., Altay A., Baykal-Grsoy M. Data-driven predictive maintenance scheduling policies for railways. <>/ExtGState<>/XObject<>/ProcSet[/PDF/Text/ImageB/ImageC/ImageI] >>/Annots[ 17 0 R] /MediaBox[ 0 0 595.32 841.92] /Contents 4 0 R/Group<>/Tabs/S/StructParents 0>> Therefore Macro F1-Score and Confusion Matrix is used as performance metrics. and transmitted securely. It furthermore presents the objective of applying the algorithm and what data was used. These are the parameters considered within ANN. The findings of the paper result that this hybrid model relying on expert knowledge and machine learning theory increases the overall accuracy than they could individually. Bethesda, MD 20894, Web Policies Data contains readings from 52 sensors and corresponding time stamp. [67] investigates a degradation model on bearing performance by applying LSTM RNN. A web application is developed using HTML and Flask web framework which is capable of predicting single point input and multiple inputs. For selecting best features, distribution of sensor readings corresponding to Noraml class and Broken class is plotted. Sakib N., Wuest T. Challenges and Opportunities of Condition-based Predictive Maintenance: A Review. <> It is not an exhaustive list, but it can give insight into where more efforts should be focused. Hence, the question becomes when maintenance should be deployed. Figure 3 gives an overview of what type of approaches exists. This challenge is especially true for vibration sensors. Typically, pump failures are rare, so using a supervised machine learning model is not typically practical. These practices, however, can sometimes lead to unnecessary part replacement, maintenance costs and labor. Here literature was recommended for further reading to deepen the understanding of how model construction could be done. If a critical component fails at a bad time, it can cause severe damage and represent a significant loss. Data-driven models can be built to estimate the remaining-useful-lifetime of complex systems that would be difficult to identify by man. The user provided the analytics company with a years worth of historical data from test bed data sets and sensors on the piston, suction and discharge mechanisms on two pumps in the field. This paper presented a literature review on current state-of-the-art applications within the field of predictive maintenance. A set of challenges were identified, during the literature review. A feedwater pump failing results in a large capacity reduction or full shut-down depending on the system setup, which in turn presents an opportunity loss and safety issue for asset operators. Parameter Tuning allows for optimising the accuracy of a model. WFE is a local mean of logarithmic vibration energy. It further allowed for identifying certain trends within the predictive maintenance field. Chen J., Chen W., Huang C., Huang S., Chen A. Learn the ins and outs of how AODD pumps work in this webinar. Notification of changes will be made as quickly as possible; please keep this in mind when arranging travel, as SPE is not responsible for any fees charged for cancelling or changing travel arrangements. Therefore, predicting failures in advance and pinpointing the source unlocks significant value. Furthermore, given a little pre-processing the degradation curve becomes fairly detectable. ; writing, J.F.O. x=6?eM/3cOv}PFH|AwFJ$wzrT&^lb6 >8yq~0WISv7l2m??EQK$(8,`3{~8qaA(!O /0Og&44Jf'>y7b}&-0$2,MA*8 D +7>-olr&%ik1OVs/"j'FSV y89~7P0dq:4>e l+ This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (, machine learning, predictive maintenance, remaining useful lifetime, state of the art review, Uckun S., Goebel K., Lucas P.J. Tinga T., Loendersloot R. Physical Model-Based Prognostics and Health Monitoring to Enable Predictive Maintenance. Small team is taking care of water pumps in a small area far from a big town. This step is repeated for all the sensors and selected sensor_00, sensor_04, sensor_06, sensor_07, sensor_08, sensor_09, sensor_10, sensor_11 and sensor_12 as final features. These failures cause huge problems to many people and also in some cases serious injuries. Another common acknowledgment is that the individual case study requires a different approach. Fernandez-Delgado M., Cernadas E., Barro S., Amorim D. Do we Need Hundreds of Classifiers to Solve Real World Classification Problems? Vibrations are common within pumps, but also generally applied. Hybrid models combine two approaches and hence utilises the strength of each method to overcome the weaknesses of the other [23]. In case of sensor_01 there is lot of overlap between distribution of Normal state and Broken state so Normal state and Broken state cannot be easily seperated. Pumps plays an essential role in various industries and as such clever maintenance can ensure cost reductions and high availability. To improve the accuracy of the prediction, BS-SRCKF is deployed for the state of charge, which in turn is combined with MHKF and an EKF to do a joint estimate of the state of charge and state of health. 110. Wang J., Zhang L., Zheng Y., Wang K. Adaptive prognosis of centrifugal pump under variable operating conditions. 15. Another method of overcoming data reflecting a flexible operation is presented by Moleda et al. A trend found within the data sets utilised is that many rely on vibration measurements [81,82,83,84,86,87]. Learn more The difficulty of creating models on pumps and CHPs stems from their flexible nature. The paper sets out to develop a dimensionless indicator. In the paper by Lee et al. Learn how to use physics-based models to build digital twins and apply machine learning to predict failure. ; All authors have read and agreed to the published version of the manuscript. Ruiz-Sarmiento et al. As CHPs are composed of various components, it becomes tricky to create just a single model for the full CHP, but requires a model on component level. Despite this, efforts have been invested in developing a structured approach to create and streamline model construction. This is done in the work by Cao et al. Hydraulic, mechanical, and other failures. The last two steps consider deploying and maintaining the model. 1 0 obj In the oil and gas industry, the uptime of industrial pumps is especially critical. There are three basic types of pumps, and they are classified by how they transport fluid: positive-displacement, centrifugal and axial-flow. The downside is that it is a black box, meaning low interpretability. 1215 September 2017; pp. An example of a common pumping system is seen in CHPs. Yu Y., Hu C., Si X., Zheng J., Zhang J. In a paper presented by Silva and Capretz [38], two fans are studied by applying CNN. Furthermore, a component can have multiple sensors as a source for a model, which requires the developer to align and pre-process data [86,87]. A comparison between applying LS-SVM as a static and dynamic version is done for presenting the results. The method is verified on a battery data set provided by NASA. Then literature with applications in various areas were introduced, this allowed for the identification of advantages and limitations of certain algorithms. [37] finds that ELM has a fast iterative tuning process for deciding on hidden layer parameters. [57] initially applies an unsupervised clustering algorithm, KSC, to initially identify anomalies within the data set, before applying the LS-SVM regression for prognostics. [. Pumps are vital to industries including water treatment and wastewater facilities, power generation, oil and gas, food processing and more. The class prediction and corresponding probability values can be downloaded as a csv file. Predictive Maintenance Solutions for Pump Applications, Podcast: Pump Industry Insight with Bob Domkowski of Xylem, The Current State of the U.S. Pump Industry, Software Helps Water Utility Avoid Unplanned Downtime, Basics of AODD Pumps & the Technical Advantages, Vacuum Technology Fundamentals & Innovations, Increasing the Lifespan & Reliability of Electronic Components, Not All Predictive Maintenance Systems Are Created Equal, In-depth articles on pump industry issues, Expert insights into important topics in the field. [83] presents a model, which initially clusters the operation modes of a centrifugal pump by applying GMM Clustering before estimating a RUL with a PF. Son et al. As industrial process might have several similar or identical components, it is important to investigate the scale-ability of developed models for them to be used for plug-and-play. Within the field of PHM, Predictive Maintenance, PdM, has been gaining increased attention. Structured PdM literature review of 30 papers and their algorithms. This refers to that no single algorithm is superior to the other, as they each serve a purpose and is case dependent. Hence, power grid operators depend on the availability and quality of the product provided by the CHPs, while CHP units strive to stay competitive. Criticality analysis is essential in order to select the pumps for which predictive maintenance solutions can best be applied and to choose a solution that can provide the most value. Sensor_15 is dropped from the dataset as all of its values are missing and for all other sensors most of the values are missing in Broken state so filled with out of distribution value, -1. Literature on pumps are more commonplace as they serve a general purpose, but it is in most literature limited to fault detection and diagnostic. Here, some models present themselves more lucrative, as they do not require a large amount of pre-processing or small amounts of domain knowledge. If vibration data is not available other methods have to be deployed. Selected time window of 10 minute for prediction and data is pre-processed by shifting the labels by 10 minute. Sampaio et al. In regards to the pump and CHP domain, a challenge was identified concerning the operation. The focal point of this paper is within data-driven models, but the authors saw the need to mention other applicable approaches and recommends the interested reader to consider the following book by Lughofer and Sayed-Mouchaweh [24]. In the above plot blue line represents sensor reading variation with time, green line represents machine status variation with time and red dots represents missing values.