Predictive maintenance for TBM's

ACCIONA´s most recent experiences in the construction of Tunnels & Metro projects has allowed ACCIONA to make huge strides in predictive analytics. Using collected data from the tunnel-boring machines (TBM) that dug the tunnels, our experts developed an algorithm that allows the Company to predict when a TBM’s drive motor may experience trouble – before it happens.

The world is saturated with data. The challenge lies in making this data work for us. In the construction industry, digitization is advancing quickly – from BIM technology to the sensors in protective gears and machinery, and to the drones collecting data from construction sites. Access to real time information has delivered enormous benefits to our industry, in terms of efficiency, cost control and health and safety at work.

Collecting real time data is one thing; analysing it to predict the future involves a different order of complexity. Yet the predictive power of data analytics could be a real game-changer for our industry.

The experience gained by ACCIONA in the most recent tunnels & metros projects around the world such as the Metro in Quito (Ecuador), the Follo Line (Norway), the Legacy Way (Australia) and, most recently the Dubai Metro, has given ACCIONA the opportunity to develop and test its predictive analytics tool for Tunnel-Boring Machines (TBM) .

For example, at the RTA´s Dubai Metro Red Line extension, the Civil Works JV formed by ACCIONA (leader) and Gulermak (Turkey) installed a Tunnel Control Centre (TCC) developed by the Technological Innovation Direction of ACCIONA Construction S.A.

At the TCC, we were able to remotely monitor (e.g. every 2 seconds) and analyse in real time  all the data sources provided by the TBM. In this case, the TBM used in Dubai, had more than 400 sensors, providing a volume of data greater than 18Mill for each day of excavation. The TBM that bored through the hard sand to connect the residential area of Discovery Gardens with the industrial area of Dubai Investment Park completed the 3.2 kms of tunnel in a record 7 months –with a maximum advance rate of 37m in one single day–  thanks to an impressive availability rate of 70%. However, in these kind of mega projects, even small periods of downtime have a heavy burden on the job efficiency. Hence, we endeavoured to improve the efficiency of TBMs even further, by making sense of the staggering volumes of data these machines were generating every day. We all know that delays cost money, they  endanger deadlines and budgets, and upset clients. Fortunately, this did not happen at the Dubai Metro, where the CWJV managed to finalize the tunnel boring several days ahead of the scheduled program.

The fact that a TBM is a mechanized machine on wheels makes it a good candidate for predictive analytics. The repetitive nature of the tunnelling works –even when the rock and external conditions change –makes it suitable to collect and classify data and search for patterns in the machine’s performance and learn from them. It allows you to collect very large samples of the same task, isolate periods when problems arise, and compare them with periods of normal operations.  

A TBM is like a moving factory - one of the largest mechanised creatures on earth. The one deployed at the Dubai Metro Project was as tall as a three-storey building, as long as five buses (>100m) and weighed as much as 15 Boeing 787s airplanes. It also counted on a special emergency safety survival chamber for the operators.  It was fitted with hundreds of sensors, which measured 2,000 variables in real time, including the thrust, pressure, speed and performance of themotors, consumption of power and lubricants, and external variables such as the hardness of the rock or the heat and humidity inside the tunnel. Access to real time data is crucial for optimising the performance of our TBMs. We then asked ourselves: Could we use the data to anticipate potential problems and improve performance even further?

After tunnelling through 70% of the project, our TBM had generated 200 Giga bits of data – enough to fill half a million books – or the entire Dubai public library. In the face of this overwhelming amount of information, the challenge was to dig for the data that was relevant for predicting the performance of tunnel boring machines.

We worked as a multidisciplinary team, enlisting our tunnelling engineers to interpret the data, data scientists to analyze the data and store and retrieve it easily, and mathematicians who could use the data to construct models capable of predicting future performance.

The TBM stopped for short periods many times a day. That is normal. What we were searching for were something different on the very few unscheduled stops.  So, we focused on what the data was telling us prior to a stop. We examined a number of variables, and finally alighted on one factor that appeared to have a strong correlation with any unscheduled stop. With the right data sets identified, our mathematicians designed an algorithm that could predict some possible failures including a potential main drive motor failure. The model was successful in predicting any factor that could affect the motor with an 80% degree of accuracy.

The real value of a tool such as the Tunnel Control Centre that we installed for the 2. 3km tunnel at the Dubai Metro lies in the ability to apply “Big Data” techniques, such as machine learning or data analytics in order to optimize the operation of TBMs for future projects.

Surely, ACCIONA will execute new TBM projects in the Middle East and abroad and will be able to count on the experience and power of data obtained in these recent tunnel & metro projects. At the moment, ACCIONA has currently a database of more than 150 km, both in rock and soil tunnels, with the most representative projects carried out in recent years being Pajares Tunnel (Spain), Follo Line (Norway), Dubai Metro Project (UAE), Metro Quito (Ecuador), etc. Thanks to these data, it will be possible in the future to improve the drilling performance of the machines by increasing their penetration, reducing consumption and even predicting critical machine failures before they occur.

Big Data in construction is here to stay. Already, it is transforming our industry for the better. It is helping us to predict costs more accurately, schedule and plan better, and lower project risk. Even small improvements in accuracy and efficiency can make a big difference to the bottom line. But as our quest for a predictive maintenance algorithm shows, only the right data – the relevant data – is what counts. The right data can generate insights that ultimately lead to better, information-driven decision making in construction projects.