Internet of Things (IoT) approach for predictive maintenance: a manufacturing case study

22/07/21

< Project overview >

Introduction

In this project, we have developed and demonstrated an IoT-based Predictive Maintenance (PdM) solution and installed it on a legacy manufacturing process from an SME manufacturer. In this scenario, a suite of digital sensors continuously monitor and report on the working condition of equipment; the sensor data is used by the trained machine learning model to predict breakdowns before they actually happen; and this information can then be used to better inform maintenance schedules.

The research team from the Department of Automatic Control and Systems Engineering (ACSE) at the University of Sheffield has developed an industrial IoT solution and introduced a PdM workflow.

Tinsley Bridge, a prestigious SME manufacturer in the Sheffield steel industry, has provided a legacy manufacturing facility as a case study and tested the developed IoT system on the shop floor. Siemens also supported this project by providing access to their cloud-based IoT platform-Mindsphere and their Industry Edge solutions.

Project lead

Professor Ashutosh Tiwari, Airbus/RAEng Research Chair in Digitisation for Manufacturing, Department of Automatic Control and Systems Engineering (ACSE), the University of Sheffield.
Dr Boyang Song, Research Associate, Department of Automatic Control and Systems Engineering (ACSE), the University of Sheffield.
Alex Kelly, IT manager, Tinsley Bridge.

Project aims

Many SME manufacturers are still operating large sets of machines and equipment without built-in digital connectivity. These legacy but durable manufacturing facilities may remain as valuable assets for the business: they will not be replaced in the short term. However, the maintenance cost and the breakdown risk could be substantially reduced by implementing the PdM approach, which is permitted by advances in IoT and machine learning technologies.

Both academia and industry are interested in new technical solutions – but there remains a sizeable gap between theory and practice. There are plenty of laboratory studies and prototyping. However, very few end-to-end solutions with IoT and machine learning technologies are being tested in production conditions on the factory floor.

This project aimed to introduce and demonstrate emerging IoT technologies to industry. It is anticipated that the project will tackle the following Pitch-In recognised IoT barriers associated with access to data, knowledge exchange, and the implementation of IoT, sensor and digitalisation technologies on existing processes:

Resource access barriers: we will address this challenge by interviewing the shop-floor operations staff and business management board members to define the IoT use case. Then, based on the use cases, we will identify the required hardware and software and their associated costs required to implement the IoT solution.
Dispositional: we will address this challenge by installing an IoT solution on the actual production line as an example, to provide insights and prove business added value for shop floor staff, engineering managers and the business decision-makers.
Business case: we will look to address this challenge through the deployment of IoT on a manufacturing process and developing an IoT business case demonstrator for a broader range of SME manufacturers.

What was done?

The main activities this project focussed on were:

Based on a case study of a legacy manufacturing facility provided by our industry partner Tinsley Bridge, we have identified the priority digitalisation opportunity via interviewing the manufacturing managers and learning about manufacturing processes and control policies.
We have developed an IoT system with industrial digital sensors in the University of Sheffield’s lab and deployed them on the Tinsley Bridge’s shop floor where no network was installed on the production line prior to this project.
We have applied our developed IoT system to monitor production processes and equipment in Tinsley Bridge’s production line, thus translating the manual inspection activities into digitalised ones.
We have developed machine learning/deep learning methods through a collaboration between the University of Sheffield and the University of Cambridge (Pitch-In project M12), to analyse the IoT system data and so predict the process conditions. With this new information, Tinsley Bridge’s maintenance team, the target end user, can better schedule and prepare maintenance tasks before failure happens and so better utilise their equipment.
Under the Covid-19 lockdown conditions, we have connected our IoT system with the internet so that Tinsley Bridge’s maintenance manager can access the states of the manufacturing process and equipment remotely 24/7. The IoT system can send out timely alerts via email or SMS messages.
A portable desktop-size demonstrator, representing a simplified PdM workflow, has been designed by the University of Sheffield, to transfer the knowledge and experience gained from this project.

Results

We have developed an IoT framework including hardware and software solutions targeting average SME manufacturers as end-users. The proposed approach has minimal infrastructure prerequisites to allow building up from zero digitised legacy facilities.

The sensor and edge hardware selection has balanced a consideration of, on the one hand, the desire for a low cost solution and, on the other, the need for sufficient reliability to survive in a harsh industrial environment.

The dataset collected from this project provides empirical data of manufacturing process quality and equipment health conditions.

It reflects the data quality and imbalance issues commonly encountered in industry datasets, and which can be challenging for any attempt to apply machine learning/deep learning directly. The dataset has been normalised and published in order to let more researchers explore these industrial challenges.

The dataset and associated problems have been shared with another Pitch-In project (M12), a collaboration between the University of Sheffield and the University of Cambridge. A deep transfer learning approach has been developed to address the data quality and imbalance problems for PdM applications. A paper presenting this work and results has been submitted to IEEE Systems Journal.

Deliverables and other tangible outputs

This project has delivered a business case report for Tinsley Bridge, a typical SME manufacturer, based on the interview with Tinsley Bridge’s manufacturing staff. The report covers the state of the manufacturing facility, the digital transformation opportunities, required costs and potential benefits for IoT implementation on existing manufacturing processes.

This project has successfully delivered an IoT solution, including digital sensors, and industrial IoT hardware and software, and which has been installed on Tinsley Bridge’s shop floor. It has been used and validated by the IT manager, process control manager, maintenance manager and shop-floor operators.

Guest lecture:

The experience of this project has been introduced as one guest lecture contributing to the new Industry 4.0 module AMR31001 at the AMRC Training Centre, University of Sheffield.

Song, B and Kelly, A (2020) ‘Applying IIoT and AI on a legacy manufacturing facility’, 17 September.

Publication:

Song, B, Brintrup, A, Turner, C, Kelly, A and Tiwari, A (2021) ‘Deep transfer learning for Industry Internet of Things based predictive maintenance. IEEE Systems Journal (accepted).

Public dataset repository:

Kelly, A and Song, B (2020) ‘E-coating ultrafiltration maintenance dataset’, Kaggle. doi: 10.34740/KAGGLE/DS/889404

Presentations:

19/06/2020, Research Seminar with Manufacturing Analytics Group, University of Cambridge
19/11/2020, Pitch-In Manufacturing webinar, Promoting Internet of Things collaborations

Between HEI and industry:

21/11/2020, EPSRC NetworkPlus in Digitalised Surface Manufacturing
16/06/2021, Pitch-In conference

Hardware demonstrator:

We are building a desktop size demonstrator to represent an end-to-end workflow of IoT based predictive maintenance, integrating industry control systems and edge AI applications.

Press articles:

Developing an automated industry IoT solution for predictive maintenance

Building strategic partnerships with industry: Tinsley Bridge and the University of Sheffield

Impact

In this project, we have accomplished the deployment of our designed IoT system in a legacy manufacturing facility, which has impacted and changed the manufacturing process control and maintenance activities. The IoT system has enabled the process control, which initially relied on manual inspections three times a day, to be transferred to 24/7 remote monitoring. This solution has improved inspection quality and reduced maintenance labour cost.

The relationship between Tinsley and the University of Sheffield continues to evolve as they explore the potential of new technologies together. Sharing these experiences with other SMEs and academic groups can help to build a supportive ecosystem and ultimately a creative region known for its innovative practices.

The research partnership between the University of Sheffield and the University of Cambridge has also been enhanced in the field of IoT and data-driven manufacturing analysis technology.

The desktop-size demonstration platform is a precious legacy from this Pitch-In project, and will be presented in dissemination activities with a broader audience from the research and industry communities once the pandemic permits. Furthermore, we will continue to develop it as a testbed and the grounds for establishing new research projects in digital manufacturing themes.

Next steps

Finalise the building of the desktop-size demonstration platform.
Disseminate the project outcomes in industry workshops and academic conferences.
Seek opportunities to extend the current industry-academic research partnership and develop new research proposals targeting, for example, the EPSRC Made Smarter theme.

Lessons learned

The collaboration between Tinsley Bridge and the University of Sheffield provided the researcher with a priceless opportunity to walk around the shop floor and communicate with staff fulfilling various functions within the manufacturing business. In this manner, the researcher has gained a solid understanding of industrial problems and explored the possible use cases and applications of emerging technology. It also provided a unique opportunity to validate lab research in the production environment.

While the pandemic disrupted the project, it also provided an opportunity to explore novel ways of online working.

The Covid-19 outbreak has limited the physical visits to our lab, to the industry partner’s shop floor and to the academic partner’s site. While these visits would undoubtedly have helped, the disruption led to utilising new ways of working to respond to these unprecedented times, such as using GitHub for remote collaborative programming.

We have transferred many testing tasks to virtual commissioning via simulation methods. A further consequence of the pandemic is that the global microchip shortage has heavily impacted hardware acquisition for our demonstration platform.

If a suitable data-sharing policy or platform had been set up within the Pitch-In cohort, it would not only allow partners within a single project to share data more securely and effectively but also enable interdisciplinary experts within the Pitch-In cohort to collaborate with controllable access to the dataset.

Access to resources, deliverables and media content

Public dataset repository:

Kelly, A and Song, B (2020) ‘E-coating ultrafiltration maintenance dataset’, Kaggle. doi: 10.34740/KAGGLE/DS/889404

Presentations:

19/11/2020, Pitch-In manufacturing webinar, promoting Internet of Things collaborations between HEI and industry

16/06/2021, Pitch-In conference

Press articles:

Pitch-In interview: Developing an automated industry IoT solution for predictive maintenance

Building strategic partnerships with industry: Tinsley Bridge and the University of Sheffield

Figure 1: The digital transformation of manufacturing process inspection (left: original manual inspection; right: installed IoT solution).

Figure 2: Dashboard of deployed Industrial IoT system, monitoring manufacturing process conditions and IoT hardware and software status.

What has Pitch-In done for you?

Our Pitch-In project has been an extraordinary journey since we started collectively proposing our project.

The Pitch-In team has provided comprehensive feedback and uninterrupted support, even during the disruptions of the epidemic. Thanks to Pitch-In’s sponsorship, we have successfully developed our IoT research, and tested and applied it on the shop floor with the industrial partners (Tinsley Bridge, Siemens).

A new academic partner (Cambridge University), introduced from the Pitch-In cohort, later cooperated on multiple projects.

The Pitch-In research webinars and network events helped us to expand our research connections and impact a broader community. We especially appreciate the additional support received from the Pitch-In to accelerate the development of the demonstration platform while we met the challenge from the Covid-19 lockdown.