Transforming Tunnel Coordination and Constructability with Computational BIM

SAGAR THORAT: Good morning, good afternoon, and good evening to everybody that's watching today and that's joined us. Hope you had a great AU round. So our topic today is "Transforming Tunnel Coordination and Constructability with Computational BIM."

And I'm glad to be with my colleague Thinagar Ramalingam from Gamuda, Singapore. He works there as a BIM manager. And myself, I'm Sagar Thorat. I'm the senior technical solutions executive here at Autodesk ASEAN, based in Singapore.

So a brief introduction about the presenter himself. So Thinager is a masters in BIM management and construction field, and graduate of civil engineering. He has more than seven years of experience in BIM on international projects. He is skilled in creating and managing BIM models across multi-disciplines, including structural, architectural, and M&E. And he is proficient in coordinated combined services models for complex projects, such as buildings, tunnels, viaducts, and bridges.

It's been my privilege to work with the Gamuda team for over the last two years, and being part of their digital transformation. And colleagues like Thinager have made it a pleasure. So over to you, Thinager.

THINAGAR RAMALINGAM: Thanks, Sagar, for giving a brief introduction about me. Hello, everyone. I'm Thina.

Let me go through about my company, which is Gamuda Berhad. Our company was established in 1976. Our main mission for our company is still reliably deliver innovative world class infrastructure and premier lifestyle properties for our customers.

Our vision is we aim to lead the region in innovative breakthrough solutions for large scale public infrastructure and property development. Our strategy is to achieve sustainable growth through our 3C approach. One is capacity, capability, and competitiveness.

Our core business is engineering and construction, property development, and infrastructure concessions. Our global presence spanning from United Kingdom, Bahrain, Qatar, India, Vietnam, Malaysia, Singapore, Taiwan, and Australia. And this was a brief introduction about my company.

The main agenda of this two-day conference is tunnel design, tunnel center versus track center, computational BIM workflow, predictive scripting, and key takeaways. Let me go through the tunnel design.

The image, what you see on the right-hand side here, is a tunnel, a typical tunnel ring design for our project in Singapore. What is CRL? CRL is a Cross Island Line project, which is launched by our client, the Singapore government. It's about 50 kilometers of underground tunnel, one of the largest underground tunnels in Singapore. It's spanning around 50 kilometers.

And the ring, what you're seeing here, is a typical tunnel ring for this CRL project. This ring comprises of eight segments. You can see this S1L is a key segment. And this S2L and S3L is a counter key segment. And remaining, we have five typical segments.

The thing why we have five typical and three non-typical segments is because of the key segment. You can see the top view of the three typical segments. One is key segment, five typical segments, and two counter key segments.

The key segment position would not be same at every adjacent ring. You can see, if you see, this magenta color segment is the key segment. During tunnel boring, we cannot install this key segment at every same position. We need to install into the counter active position.

This is to counter when the tunnel is going to the left. We have to install the-- when the tunnel is coming to the left, we need to counteract the force towards the right. When the tunnel is turning to the right, we need to counteract the force towards left. So we have to install the key segment at the right if the tunnel is turning to the left.

So it's just straightforward for all. If it is moving towards the up, we need to install the key segment at down. And if it is moving towards the down, we need to install the key segment at towards the up.

So we have six possible key segments for right-hand side and six possible key segments for left-hand side. The right-hand side possible key segment is starting from 12 o'clock. This is key position number one. And we need to install until this is the possible key position until 3:00 PM. Then for left-hand side key segments, spanning from 12 o'clock to 9 o'clock. So this is left possible key segment and right possible key segment.

And now we come to the details of inside the tunnel ring. So basically, the tunnel ring is not just a precast segment. Inside, the tunnel ring's got minor dimple holes, which is useful for our services bracket coordination. You can see what I highlighted in the yellow color is the dimple hole. Each tunnel ring comprises of 66 set of possible dimple holes as a whole. For clear understanding, I just expand the tunnel ring into the plan layout. If you count all these small holes, it will be 66 numbers in each tunnel ring.

| radial distance between the two dimple hole will be 275 mm. The maximum radial distance between the adjacent segment, if I calculate these two dimensions, it will be 275 mm. If I calculate the maximum distance between the adjacent segment ring, this is the connection point of the two segments. It will be 828 mm. So the typical is 275, and the maximum is 828 mm for the dimple hole location.

Why we need to be so particular on dimple hole? Because we need to mount our services, all the services. Inside the tunnel, we need to have services, like signaling comms and other [INAUDIBLE], drainage cable, et cetera. We need to mount the bracket for the services cable. So all the bracket will need to be only installed and drilled at the dimple hole location. That's the coordination. That's the bigger coordination we need to do for services point of view.

So if you see our typical section here, the brackets, A, B, C, everything we coordinated at the dimple hole location. The lines, what we've shown in the magenta color, is the line. What I've shown here in plan as a hole. So it's just a typical section. We have 66 number of dimple holes as shown in the section here.

Why we get the restriction on drilling the dimple hole only at dimple hole location, drilling the hole only at the dimple hole location? Because these restrictions are placed to avoid hitting any segment of Rebar and to preserve the waterproofing integrity of the tunnel segment.

The tunnel segment itself, the consultant, they design in such a way that they avoid the Rebar at dimple hole location for us to provide any drilling. Because it's an underground tunnel, it's very sensitive for waterproofing, integrity, everything. So we must need to make sure that we are not hitting any Rebar while drilling for the bracket position.

So the direction of drilling must be from center of tunnel to the center of dimple hole. Or in another word, the drilling direction must be perpendicular to the direction of the dimple hole center. So I show here the sample how we are drilling at the dimple hole location.

Basically, before the current project, which is CRL in my previous project, we did this manually. At that time, we really not like that much expert, and do all these kinds of remodeling. So that time we did the manual way of coordination. What we did is same like what I explained previously here.

We extend the ring in plan layout. And we plot the dimple hole positions. Then the key segment. Basically, the key segment position, our tunnel team will provide to us. Once the TBM machine starts rigging into the ground, based on the ground condition, they will position the key segment, and they will give us, the build team or coordination team, this is for first ring. This is your key position. It will be 1 o'clock. For the second ring, your key position is 9 o'clock, something like that. They give us the angle of rotation.

So what we develop-- in current way of coordination, we develop the model based on their info, but in previous way, we just do it in manual way. We plot everything in 2D. And if you see the brackets here, you can see this rectangular box is the brackets.

We coordinated at this dimple hole, dimple hole 1 and 2, this dimple hole location. Then the second bracket is coordinated for this location. If we go to the next ring, we cannot get this dimple hole at this point because the key segment from here and the key segment here is different.

Like I explained before, the key segment is not the same at every adjacent ring. So we cannot get the possibility of a dimple hole in this segment. So we need to extend the bracket to tally with this dimple hole, and we need to extend the bracket to tally with this dimple hole so that only for that particular location we need to have a longer bracket.

Same like in this way, our project is spending around 1 kilometer, which comprises of two twin-bored tunnel. So if we do this manually in this way, it will take lot of time to prepare this 2D, and to coordinate this bracketing. That's why we develop everything in Revit Dynamo, using Revit Dynamo. Even Revit, also, if we model the family, and if we plot in the segment, and rings one by one manually, it will take lot of time. That's how Dynamo theory will be very helpful for me to run the rings throughout the whole alignment.

This is a sample of how we're coordinating the brackets inside the tunnel ring together with the dimple hole location. Let-- I'll explain what is tunnel center and track center. So when-- this is a typical bored tunnel section. When client sent us the track, when client sent us the alignment track, they usually send us the alignment for track center point.

So consultant also will develop their model based on the track center point. You can see this point 1, and this point 2 is track center point. And this point 3 and point 4 is tunnel center point. This point 3 and point 4, I added by myself, because the reason why I added the tunnel center point, we need to develop our family based on the angle of rotation towards the center.

So the current model, what consultant provided is without any parameter. It's just conceptual mode. Even there is no segmentation. And also, there is no dimple hole position.

So using this model, which they develop based on the track center point. I need to acquire the design tunnel center point. How I acquire the design tunnel center point is the big question for me. And I explore lot of options. And Dynamo is the one that helps me to get the coordinates.

This is the sample of my tunnel ring family, which I developed by myself. We can say it is contractual model. And this one, we can say as consultant model. The contractor model is mostly based on construction. This one is just a conceptual design model.

So our construction model, if you see the point, it's based on the tunnel center point. The design center of the tunnel and track center point x-axis, that will be a 200 mm offset, and in z-axis, it will be 1,755 mm offset. So if we want to develop our own family, we need to-- our tunneling family is-- we need to design by parametrically. We need to rotate from the angle of rotation at the center.

So I show you the video on how we set the parameter of rotation. I'll play the video. This is my family. If I key in the value here at the parameter, if, let's say, I enter the value of 0, the whole tunnel segment itself will rotate together with the whole set. So it will be very easy for us to glean the angle of rotation once the tunnel team gives us the information.

We can derive the tunnel center coordinates mathematically if the bored tunnel is straight. So our ultimate aim is we need to derive the tunnel center point for our family to run inside the alignment-- run in the alignment. So if, let's SAY the tunnel ring is straight without any elevations, or any level difference, we can make the coordinates. We can calculate the coordinates manually or mathematically by detecting x value and y value.

For example, if you see from top view, you can see the difference between the point 2 and point 3 will be 200 mm. If the tunnel is straight, it's exactly 200 mm. But we cannot derive the coordinates mathematically if the bored tunnel has different in elevation in alignment.

For example, if there is a difference in alignment, the first point, I play this video. The point 4 and point 1 will be the lowest point, and the point 3 and point 2 will be the highest point. So you cannot see-- you can see clearly these two points are in different positions. So we cannot use any mathematical formula or derivation to acquire that coordinate. So the only way is we need to get the coordinate using Revit Dynamo.

Computational BIM workflow-- how I get that coordinates using that Adaptive family. I'm going to explain. Basically, using Civil 3D, consultants will send us the alignment drawing. Then the BIM team will use that alignment drawing, and export the xyz coordinates of the track center point, and run the track center point using Dynamo script. Then model that in Revit.

In either way, we need to acquire that coordinates from that family. So currently, the model, what we modeled, is based on track center point. We already get the model based on track center point, but we need to have a center point of tunnel for us to rotate the family parametrically.

So we need to get, again, input from that family. So we need to add two more points at the center point. Then we need to export the coordinates into Excel using Revit Dynamo.

Acquiring the coordinates from Adaptive family. The first step of the Dynamo script is we need to acquire the coordinates. Basically, this is the consultant model. They model based on the track center point, center point 1 and center point 2.

Now, I manually added point 3 and point 4 to acquire these coordinates. So basically to determine the output we need, we must first identify the primary node. Once we have established the primary node, we can develop the necessary input for it based on its specific requirement.

So in our process, our ultimate scope is we need to get this xyz. This is the output for our Dynamo script. So we need to get the xyz of tunnel center point. So this is the output. This is one of the primary nodes.

Then another primary node is how we get that xyz coordinate using this adaptive family, which is modeled by our consultant. So we use their own family to acquire the coordinate. So the second key node is adaptive component by location.

Because you can see we need to get this third point only, we don't need another point. If we get this third point for next adjacent tunnel, this is the third point. For next adjacent point, the next one is the third point. So if we get all the rings' third point, we can get xyz coordinates easily for throughout the whole alignment.

So the third key segment is list get item at index. So we need to get only third item at the index. So using the first key node, adaptive component location, we need to load all the elements of family so that we can get all the coordinates. So the family type must be-- we need to properly select the tunnel segment family, what model by consultant.

Once we gain all these inputs to the process, this is-- once we gain all this input, this is the output we get, the xyz of this ring, then XYZ of each ring. We get xyz of each ring. I mean, each ring comprises of four points-- 0, 1, 2, 3. All these four points we get for every ring already.

But we only need the third point. That's how we need to process it. So we transpose this whole point. Then once we transpose, we need to get the item at index. We need to get only the third item at the index. If we get the third item and index, and if we run the Dynamo script, you can get xyz coordinates of all the tunnel ring third point.

So here we go. We get xyz coordinates for all the third point. Then we can-- now we get the xyz here. How we need to export it to any format? We need to export it to CSV or Excel. Currently, I use data export to CSV.

So I key in this item to the data, and we located the file path so that the CSV file is saved at some location. If I run the data, this is the final output. xyz, we already acquired from the consultant tunnel ring family.

Now, how are we going to run that? Run our family contractor ring family using the coordinates. So we need to develop another Dynamo script. Now, we could get the coordinates. And let's run the bored tunnel ring family into the tunnel alignment using the coordinates.

So here, the key node is-- first, we need to always-- like I said before, we need to always look at the output. The output here we need is adopting component by points. Now we're going to run this two-point adaptive family using the points, what we acquired from the previous script.

This is the primary node for the output. Then the process, how we get the points, we need to get the xyz points. So the key note for the process is xyz. Then how we get the xyz? Using the CXL file, what we exported in the previous script. So this is the output from the previous script.

We key in this xyz from the file path and read this Excel file. Then we have to enter the sheet name exactly what we documented in the Excel file before. For the eastbound, I just put EB. So I just link this EB to sheet name.

Then we get the Excel file. Then we get these xyz coordinates here. We already read it, but we need to list it in xyz separately. So we need to get the xyz. If you see here, this column in Excel, we call it as 0 column. And this one is 1 column. And this C is 2 column.

So we have to get xyz separately. That's why we separately get the item at index list. Get item index for x list. Get item at index for y. And list get item at index for z.

Once we get the xyz separately, we need to list it in the way that x comma y comma z so that we are able to run the ring family into the coordinates. So the next process, we need to sublist this coordinate. Currently, it's just xyz.

All the points, we have the whole list. But our ring family is made up of two point adaptive family. So it needs to be run from 1 point to 2 point. Then second point to third point, then third point to fourth point, something like that. The point will connect one by one, and create the whole tunnel ring tunnel alignment family.

So 0, we had to arrange it in the way, that 0, dot, dot, 1. Then the offset must be every one row, every one row. Because if I make this to 0 to 3, it will be 0, 1, 2. If I make it to 0, 2, 4, it will be 0, 1, 2. It will be four points at one list.

But we only need two points at one list. That's why we put 0 to 1 range. Then the offset must be, like I explained before, 1 to 2, and 2 to 1. So the offset must be 1.

If I put two values here, it will be 0 to 1. Then it will be 1 to 2, and, again, 3 to 4. So it will not connect from 2 to 3. That's the reason we put one number here. So the offset must be 1, and the range must be 0 to 1.

Now, once we run the sublist, we can get these points, xyz, or 0 point xyz for 1 point. Then what else we need to do is we need to import the family. We need to include that family into the output node. So we included the points, and we included the exact family, what we developed as a contractor. We developed our parametric family. We input that family into the family type.

Then once we run the whole 1 kilometers of tunnel, a ring family will run into the whole element within 15 minutes. If we do it in manual way, it will take around one day to manually model. And using manual way, there is no-- even for me, I don't get how we model for the alignment. If, let's say, there is an up and down turn. So there is no other way to model it in manual way.

If you model it manually either, if it is straight, we can model it manually. But if it is an undulation, we cannot model it manually. That's how Revit Dynamo comes into the way, which is helpful for us to develop the whole complex project tunnel modeling within a few minutes.

So now I hand over to Sagar. I hope with this you can get the good knowledge, gain the better, decent knowledge of tunnel modeling using Dynamo.

SAGAR THORAT: So thank you, Thina, for wonderfully explaining how this has helped you. So let me just-- so with predictive scripting, we'll understand how machine learning and Dynamo can be leveraged. And Thinagar and Gamuda team have been leveraging on this particular feature of Dynamo for the last one or two years now. And it has enabled them to streamline, and even fasten the Dynamo graph creation.

So what is this node autocomplete? So in Dynamo, node autocomplete is machine learning empowered, so it has algorithms, and it is a subset of artificial intelligence. And it has been trained on real Dynamo graphs.

It hierarchically ranks results. And it ranks the predictions based on the probability numbers. Now, there are two types of nodes in this case. These two types of nodes are node searches, is node type match, and recommended node. Now, you should use the node type match approach when discovering possibilities within a single shelf in the library or the data sets that the machine is learning from, or more technically accurate, within the same tier, or inheritance of subjects.

And you should use the recommended approach by default, unless it doesn't offer suggestions that you desire. In time, the recommended approach will get more and more precise as we develop the program further, allowing for swifter and more accurate craft decisions.

The book methods work quite well for different scenarios, but the static node type math approach will always require more brain power than the machine learning version, as the machine learning version will get smarter and smarter over time.

Now, let's see this in action on how the Gamuda team uses it with a minor example. So here, we are here to find the family type gnome. So normally, the user won't remember what kind of node to use. But by double-clicking on the family type on the input node itself, he is able to call out the suggestions from the autocomplete node, and then you will get the family type. And now you're able to select that particular family.

So for the sake of users, I would be replaying this. Here we are calling out autocomplete. We are switching between the types, which is node type match, and recommended nodes. And we are actually now switching back to recommended node, and getting the family type node out. And this enables you to faster create the scripts or the Dynamo graphs.

Now, you might be asking me, how is the step? What is going on behind this? And technically, how does this work? So let's try to understand that. From a Dynamo graph user perspective, when the user is triggering this machine learning empowered node autocomplete service, first, there is a backend engine that actually derives these training data sets from current Dynamo users that have been applying them for years.

Then the algorithm, there are relationships that are defined by experts via Dynamo graphs. And these Dynamo graphs have been learning over the last years or so. The node relationships later are extracted into a machine learning algorithm model. And that gets transferred over. And it gets collected as a database.

Now, when the user at the front end triggers the machine learning algorithm, at that time the node autocomplete is activated. Now, mind you, internet connectivity is required for this purpose so as to gather a mixture of data sets and to be able to leverage on this functionality. You can either call it out from the input, or you can call it out from the output. In this case, the input is a closed curve, and output is a surface. You can call out the autocomplete from both ends.

After the call out, Dynamo pings the machine learning of the autocomplete service, and the machine learning will look through the triggering ports, whether it's input port or port or an output port. And it returns hierarchical results in a list. And then you can choose from this prediction at the highest percentage or the lowest, or the ones that seem correct for your application of the input or the output.

Now, these hierarchical results will list this return into the Dynamo user interface from the service. And the users can now select these results and place it in Dynamo in the Canvas, and wiring them across, and triggering the port for further data manipulation.

So with this, we would now like to summarize our presentation or our content in few key takeaways. And with the permission of Thina, I would like to summarize it from the people standpoint, where we see that improved decision making is the biggest benefit when it comes to teams working together to be able to comprehensively review 3D tunnel models, whether it is to be reviewed in the authoring software itself or on the Autodesk Construction Cloud. This helps project managers and decision makers to make an informed choice.

Secondly, over the years, Gamuda team internally has been running skill development programs, and periodically raising their bar, and going into the areas of artificial intelligence and machine learning to derive more processes through.

Now, from the process standpoint, the biggest benefit we see is the time saving in the coordination process itself. If you compare that with the traditional coordination process, which Thina has very beautifully highlighted in the earlier segment.

Constructability aspects, that cannot be ignored. We all know they can be very costly on the side. And things like simple things like dimple holes, locations can be missed out if you are using a manual way of copying and pasting brackets across when you're using your CAD software. However, there is no degree of control on your data.

THINAGAR RAMALINGAM: It is coming to the technology part, which I go through before. In traditional way, we plot the plan layout of the ring, and we mark the brackets. But when it comes to the Dynamo scripting, it reduces-- if, let's say, I want to do it in traditional way, minimum for 500 meters of tunnel, it takes around 8 to 12 hours. If I use-- the first time when I develop this Dynamo script, it takes around 3 to 5 hours using Dynamo script.

But when it comes to the predictive script right after that, only I know about the predictive scripting. It reduced the whole 3 to 5 hours to 1.5 hours. So using predictive scripting, I can save around 50 percentage of my time, even in Dynamo. So the Dynamo is already reducing 50% of the time from traditional method.

Then, again, the predictive scripting is reducing time 50 percentage of the time from Dynamo like manual way of loading it in Dynamo. That time also reducing it 50% using predictive scripting. So with this, we conclude our presentation. And we open to the floor for questions. Thank you.