Artificial Intelligence is the new disruptive innovation in condition monitoring
- pabloramirezgarcia
- Jun 18, 2021
- 12 min read
Updated: Jul 4, 2021
#Artificialintelligence (AI) and especially #Machinelearning (ML) is a technology that can be key in the further effectiveness of #ConditionBasedMaintenance, in this era of continuous digital transformation in the industry, the efficient use of data, and the development of predictive capabilities will grow the prospects of multiple organisations. Different techniques and algorithms can be used as an effective tool to predict real-time faults and help troubleshooting. As suggested by Serradilla et al. (2020) these models will increase the performance of predictive maintenance (PdM), getting accurate results, gaining knowledge from the data while contrasting with a theoretical background and domain expertise. But can it create a competitive advantage? And what are the drawbacks?
Pablo Ramirez 04/07/2021 - 12 min read
Source: (pixabay, 2021)
Silvia Peres et al. (2020) defined Industrial AI as a systematic discipline focusing on the development, validation, deployment, and maintenance of AI solutions for industrial applications with sustainable performance. ML techniques aim to capture complex relationships in the data that may be difficult to describe to introduce earlier warnings on incoming failures and to determine higher accuracy in the remaining useful life predictions of systems (Wagner, et al., 2016). In Industrial applications where the need to predict failures is a clear value, the potential is remarkable. The amount of data especially in condition monitoring databases is massive, as pointed out by Domingos (2012) the raw data can construct features from it, machine learning is an iterative process of running the learner, analysing results, modifying data and repeating. This is completely aligned with the diagnostic capabilities in the industry and can be easily leveraged in the automation of the engineering process.
The training data can build models through the identification of patterns and trends (Geng, 2017). The machine learns from examples, rather than being explicitly programmed for a particular outcome (Brynjolfsson & Mcafe, 2017). This can be useful when the prognostic capabilities in a company are not totally implemented or when the next step is to reduce human dependency.
According to Mustafa's research, expanding applications of modern information technologies and automation will bring an added advantage (Mustafa & Rahman, 2015). Machine learning is being a relevant part of the digital transformation of condition monitoring systems creating #disruptiveinnovations in the industry. Brynjolfsson & Mcafe (2017) expounded that artificial intelligence, particularly machine learning (ML) is the most important general-purpose technology of our era, especially because we can now build systems that can perform tasks on their own. Thanks to that the applications in industry are uncapped, not only because it can automate manual actions but also because it can diagnose and predict failures something that was limited to humans so far. Consequently, as suggested by Silva Peres et al. (2020) the main application areas of interest appear in energy optimization, predictive maintenance, and quality control, largely dominated by deep learning in terms of the methods employed. For these methods, data is a significant constraint and in some cases, it is a challenge to maintain and acquire big amounts of data. If the obstacle to acquire, send and store data from monitoring systems is overcome the potential of AI in the industry will be completely leveraged. In this regard Geng (2017) confirmed that the raise of big data in recent years has opened up unprecedented opportunities, rapidly changing the landscape of analytics and technology. Therefore, technological developments in hardware and software will enhance the #InternetofThings capabilities, hence driving the competitive advantage.
McKinsey Global Survey on artificial intelligence (AI) suggests that organizations are using AI as a tool for generating value. Increasingly, that value is coming in the form of revenues. AI adoption is highest in the service-development and service-operations functions, reaching up to 19% of adoption in the Predictive services. In industry there is a relevant value-added in reducing the unplanned downtime, furthermore, it is even more important having the possibility to predict failures and deploy smart maintenance, thereby avoiding catastrophic consequences. According to Silva Peres, et al. (2020) a large portion of existing applications of Industrial AI is focused on increasing machine operability and uptime by detecting possible problems before they occur. Generally, such approaches can be modelled based on the degradation severity of machine performance and the processing of multiple heterogeneous data sources. These techniques will cause a reduction in costs, in the form of effective maintenance strategies, optimisation in resources, and increase in production, consequently, it will create a relevant competitive advantage, this will be even more relevant for the fact that modern equipment becomes increasingly complex and requires longer maintenance time (Ong, et al., 2020).
IBM in the US predicted that the scale of the artificial intelligence market would reach 2,000 trillion won by 2025, and McKinsey predicted that the relevant scale would reach a higher figure of 7,000 trillion won. Therefore, this study predicts that the scale of the market for artificial intelligence will grow further in five years and takes note of the active growth (Jung Go, et al., 2020).
We can see in the following figure the advance and predictive analytics software revenue worldwide in the last years (in billion US dollars).
(Statista, 2021)
The following figure exhibits the forecast growth of the artificial intelligence software market worldwide from 2019 to 2025.
Source: (Statista, 2021)
As expounded in the report from Ransbotham et al (2020) according to the 87 percent of respondents, obtaining a competitive advantage tend to drive artificial intelligence (AI) business strategies among their organizations worldwide and according to the report from NewVantage Partners (2021) 82.7% of factors driving investments into AI and Big Data are offensive (e.g. transformation, innovation, and competitive advantage).
Predictive maintenance is clearly one of the main beneficiaries in the industry of artificial intelligence with up to 24.3 % of the cases in industrial applications (Rykov, 2019) and expected cumulative revenue of 5713.6 million US dollars between 2016 and 2025 (Statista, 2021).
Predictive maintenance is the most effective maintenance strategy, achieving overall equipment effectiveness higher than 90% (Serradilla, et al., 2020), however that effectiveness can be boosted even further thanks to the implementation of Machine learning techniques, ML applications provide some advantages which include maintenance cost reduction, repair stop reduction, machine fault reduction, spare-part life increases, increased production, inventory reduction, and many more (Çınar, et al., 2020). The features of AI give even more capabilities to the normal predictive techniques because it can anticipate failures not only based on signals of the engine but also on the occurrence of determined events as expounded by Decker de Sousa et al. (2019).
With the increase of computational power and data growth in the field of predictive maintenance, research on this area tends to focus on data-driven techniques (Serradilla, et al., 2020). The growth of capabilities and the enhancement of technologies in this area imply a relevant prospect for this field. One of the main characteristics of condition monitoring systems is the management of a large amount of data, this facilitates the creation of models that allow predicting failures based on the degradation of components. As expounded in the article by Liu (2001) artificial neural networks with enough training data are capable of modelling any function or behaviour, creating the universal approximator. This can bring up the thought for the compulsory need of precise training data, but there are also models that require few parameters and therefore little training data as shown by Spiegel et al. (2018).
There are multiple pieces of research in this field that show the advantages and prospects of the implementation of machine learning in predictive maintenance. Some examples are the work of Yiwei et al. (2017) who proposed a model-based prognostic framework, showing relevant cost savings in respect to traditional PvM or the model of Sheu et al. (2015) who proposed a long-term average cost function for major, minor and imperfect repairs, assuming that the equipment degradation follows a non-homogeneous continuous-time Markov process. Deep Learning has been used as an alternative for the estimation of the remaining useful life of equipment and ball bearings. Similarly, work has also been reported to learn the health indicators for the remaining useful life estimation of a turbofan engine by using Temporal Difference learning (Ong, et al., 2020).
It looks like everything is positive and easy in the implementation of AI in predictive maintenance. The reality is that there are multiple obstacles to the effective outcome of the deployment of these strategies.
Implementing ML-based predictive maintenance is a difficult and expensive process, especially for those companies that lack the needed resources to invest in expensive IT architectures and develop the skills to deploy the necessary activities to have an effectual algorithm capable of detecting faults in real-time and diagnose precisely the cause of the failure and recommend the solution with the estimation of the remaining life of the system (Florian, et al., 2021) (Barraza-Barraza, et al., 2014). Furthermore, although it is proved that predictive maintenance is more effective than preventive strategies some models usually neglect the economic cost associated with true/false positive/negatives (Spiegel, et al., 2018).
Selecting the most appropriate algorithm for the requirements of the maintenance strategy is a major challenge and can be a relevant setback not to choose the most reliable method, however, as expounded by Çınar et al. (2020) the combination of more than one ML models can provide better predictions. In this case, the issue can be the demand for technical resources, knowing that the bigger the speed and volume of data collection and analysis, the higher are the costs related to ML investments (Florian, et al., 2021). In this regard, and as shown earlier, the expansion of modern information technologies might diminish this obstacle.
Conversely, a relevant part of costs in condition monitoring systems is the triage of deviations, diagnosis, and notification process as confirmed by Spiegel et al. (2018), which can be clearly reduced by an ML-based strategy. Moreover, a significant aspect is the fact that some organisations are averse to transmit data, which can create a need for these kinds of companies to implement a self-learn system with actionable recommendations, as suggested by Ong et al. (2020). Thereby machine learning is a cost-effective solution to reduce maintenance costs.
It is out of the scope of this article to go into the detail of the multiple algorithms of machine learning, notwithstanding it is important to understand two main different types that underline the relevance of the human factor. Decision tree structures, as their name suggests, use decision trees as predictive models in deciding how to reach a conclusion on a particular set of data. Neural networks, on the other hand, mimic biological neural networks in estimating or approximating rules or functions which depend on a huge number of inputs and are generally unknown before discovery (Geng, 2017). The selection of the approach is not only based on the data size and quality but also on the need for the human interpretation of the results.
The problem with neural networks and deep learning models is not only the high data requirements and difficulty in the interpretations of the results but also the lack of mitigation proposals (Serradilla, et al., 2020). The role of humans in the implementation of these techniques is crucial, not only in the selection of the right algorithm but in the process of training the data and the outcome of it. As suggested by Accorsia et al. (2017) and Zhao et al. (2019) it is beneficial the partnership of both automatic data analysis techniques and human knowledge, since machine experts will be key to select the right faults and the most important data for the analysis, furthermore the pre-training and filtering of data will be essential to boost the performance of PdM models. Conversely, Serradilla et al. (2020) confirmed that these new models can achieve accuracy and create tools to perform PdM automatically, removing the dependence on manual engineering processes.
Workers are directly impacted by the outcome of an industrial revolution, it is clear that the current model will change if AI takes over human activities. According to Silva Peres et al (2020) Industrial AI is leveraged to augment human performance rather than fully replace them. However, there is no clear answer and it is imperative to include ways to assess and quantify these characteristics to take conscious business actions.
Machine learning is clearly adding value to the standard diagnosis and prognosis techniques in predictive maintenance, therefore the bond between both fields seems to be a need for the development of PdM. ML techniques are a useful tool to exploit data and use them to implement predictive models and decrease failure costs, the increase of available data and the interest of researchers around ML will favour significantly the development of PdM (Florian, et al., 2021). In spite of the diverse difficulties to implement these programmes to leverage the benefits of artificial intelligence, the advantages are numerous, there is clearly a need to deploy initiatives and plans to foster the efficient implementation of AI in organisations. As expounded by Martin, R. (2009) an organization’s value equation is the value its goods and services create for consumers relative to the cost of creating that value, therefore the success will depend on the result but also in the right implementation. The demand for data scientists and the shortage of this kind of talent is a relevant constraint for organisations. This is creating a significant challenge to have an in-house analytics function and is fostering the use of Artificial Intelligence-as-a-Service (Geng, 2017) (Sorensen, et al., 2019). Notwithstanding, the digital transformation and innovation initiatives need to come from the business leaders. As suggested by Geng (2017) the vision, business cases, and talent strategy have to be kept within the organisations for commercial reasons, but the collaborations with external entities might be helpful. Furthermore, as suggested by Brown (2009), Yoo et al. (2010), and Geng (2017), the internal implementation of design thinking, experimentation, and digital innovation can generate disruptive ideas to generate customer value and market opportunities through a viable business strategy in data analytics.
The development of Condition-Based Maintenance will be an essential strategy for companies in the Fourth Industrial Revolution. In order to lead the Condition Monitoring solutions will be key to implement efficiently artificial intelligence technologies, either with internal or hybrid approaches. Take advantage of Big data technologies enhancements and disruptive innovations will be crucial to generate differentiation and #Digitalleadership.
Accorsia, R. et al., 2017. Data Mining and Machine Learning for Condition-based Maintenance. Procedia Manufacturing, Volume 11, pp. 1153-1161.
Barraza-Barraza, D., Lim´on-Robles, J. & Beruvides, M., 2014. Opportunities and challenges in condition-based maintenance research. Annual Conference and Expo, p. 3035–3043.
Brown, T., 2009. Change By Design: How Design Thinking Transforms Organizations and Inspires Innovation. New York: Harper-Collins.
Brynjolfsson, E. & Mcafe, A., 2017. The Business of Artificial Intelligence - What it can and Cannot do for your Organization. Harvard Business Review.
Çınar, Z. et al., 2020. Machine Learning in Predictive Maintenance towards Sustainable Smart Manufacturing in Industry 4.0. Sustainability, Volume 12, p. 8211.
Decker de Sousa, L. et al., 2019. Big Data Analysis for Predictive Maintenance at the INFN-CNAF Data Center using Machine Learning Approaches. Proceedings of the XXth Conference of Open Innovations Association FRUCT, 622(25), pp. 448-451.
Domingos, P., 2012. A few useful things to know about machine learning. Communications of the ACM.
Dunphy and Stace, 1993. The strategic management of corporate change. 46(8):905-1:Human Relations.
Florian, E., Sgarbossa, F. & Zennaro, I., 2021. Machine learning-based predictive maintenance: A cost-oriented model for implementation. In International Journal of Production Economics, p. 236.
Geng, H., 2017. INTERNET OF THINGS AND DATA ANALYTICS HANDBOOK. John Wiley & Sons ed. New Jersey: Hoboken.
Jung Go, E., Moon, J. & Kim, J., 2020. Analysis of the Current and Future of the Artificial Intelligence in Financial Industry with Big Data Techniques. Global Business and Finance Review, Volume 25, pp. 102-117.
Liu, P., 2001. Approximation capabilities of multilayer feedforward regular fuzzy neural networks. Applied Mathematics, 16(1), pp. 45-57.
Martin, R., 2009. The Design of Business: Why Design Thinking is the Next Competitive Advantage. Boston, MA: Harvard Business Press.
McKinsey & Company, 2020. The State of Ai in 2020. [Online]
Available at: https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/global-survey-the-state-of-ai-in-2020 [Accessed 2021].
Mustafa, M. & Rahman, M., 2015. Financial inclusion and per capita real GSP growth across fifty US states and the district of Columbia: Evidence from panel cointegration and GMM estimates. Global Business & Finance Review, Volume 20, pp. 87-94.
NewVantage Partners, 2021. Big Data and AI Executive Survey 2021, Boston | New York | San Francisco | Raleigh
Ong, K., Niyato, D. & Yuen, C., 2020. Computer Science Statistics, Engineering Product Development, Singapore University of Technology and Design.
pixabay, 2021. Pixabay.com. [Online]
Available at: https://pixabay.com/illustrations/head-board-machine-learning-4519964/ [Accessed May 2021].
Ransbotham, S. et al., 2020. Expanding AI's Impact With Organizational Learning: ” MIT Sloan Management Review and Boston Consulting Group.
Rykov, M., 2019. IoT Analytics. [Online]
Available at: https://iot-analytics.com/the-top-10-industrial-ai-use-cases/ [Accessed May 2021].
Serradilla, O., Zugasti, E. & Zurutuza, U., 2020. Deep learning models for predictive maintenance: a survey, comparison,. Computer Science, 1(1).
Sharma, S. & Wang, X., 2020. Toward Massive Machine Type Communications in Ultra-Dense Cellular IoT Networks: Current Issues and Machine Learning-Assisted Solutions. IEEE Communications Surveys & Tutorials, 22(1), pp. 426-471.
Sheu, S., Chang, C., Chen, Y. & Zhang, Z., 2015. Optimal preventive maintenance and repair policies for multi-state systems. Reliability Engineering & System Safety, Volume 140, pp. 78-87.
Silva Peres, R., X., J., Lee, J. & Barata, J., 2020. Industrial Artificial Intelligence in Industry 4.0 - Systematic Review, Challenges and Outlook. IEEE Access, Volume 8, pp. 220121-220139.
Sorensen, C., WLG, G. & Consulting, L., 2019. Tides of disruption how to navigate business transformation. LSE.
Spiegel, S., Mueller, F., Weismann, D. & Bird, J., 2018. Cost-Sensitive Learning for Predictive Maintenance. Computer Science Statistics.
Statista, 2021. Statista. [Online]
Available at: https://0-www-statista-com.pugwash.lib.warwick.ac.uk/statistics/607960/worldwide-artificial-intelligence-market-growth/ [Accessed May 2021].
Statista, 2021. Statista. [Online]
Available at: https://0-www-statista-com.pugwash.lib.warwick.ac.uk/statistics/1172729/advanced-and-predictive-analytics-software-revenue-worldwide/ [Accessed May 2021].
Statista, 2021. Statista. [Online]
Available at: https://0-www-statista-com.pugwash.lib.warwick.ac.uk/statistics/607835/worldwide-artificial-intelligence-market-leading-use-cases/ [Accessed May 2021].
Wagner, C., Saalmann, P. & Hellingrath, B., 2016. An overview of useful data and analyzing techniques for improved multivariate diagnostics and prognostics in condition-based maintenance. Denver, Proceedings of Annual Conference of the Prognostics and Health Management Society.
Yiwei, W. et al., 2017. A cost driven predictive maintenance policy for structural airframe maintenance. Chinese Journal of Aeronautics, 30(3), pp. 1242-1257.
Yoo, Y., Henfridsson, O. & Lyytinen, K., 2010. Research commentary-The new organizing Logic of Digital Innovation: An Agenda for Information Systems. Information System Reasearch, 21(4), pp. 724-35.
Zhao, R. et al., 2019. Deep learning and its applications to machine health monitoring. Mechanical Systems and Signal Processing, pp. 213-237.
Comments