Background

Air distribution plays an important role in human thermal comfort in built environment. HVAC (Heating, Ventilation, Air-Conditioning, and Cooling) sytems are the core to ensure it. Compared with nominal capacity of core equipment, Terminal air distribution system seems not to get enough attention, especailly when it comes to BIM (Building Information Model) era.

BIM facilitates the automation in construntion, but there is a long way to go. For example, Construction Drawing Standard needs to be recieved and embraced widely, which can help the ducting generation of centralized air distribution system. Identifying the geometry and use of every space is indispensible in the ducting generation. It is not uncommon to hear that the ducting generation is labor-intensive, which means that the increasing powerful machine can liberate engineers from the “trivial” and boring work. But the ducting generation is the foundation of the automatic hydraulic calculation, so what’s our current difficulties?

Bottleneck

The potential of soft- and hard-ware firstly shines through Computer Vision(CV) and Nautre Language Process(NLP) . one of reasons we cannot ingore is data collection: the tremendous amount of textual and visional data. Back to the construction industry, it is really hard and time-consuming to put all data in our hand into some well-developed data structure, including IFC^[1] and gbXML^[2].

Automated 3D model creation (for example, a quasi-automated BIM methodology) has been researched and bullet points have been extracted, like line-text extraction^[3], which is the solid foundation of the ducting generation. However, The tendency of automated ducting generation has not come yet, at least in the construction industry. Let’s wait and see it.

How to Automate Hydraulic Calculation

an Abstract Routing

Given that we get the abstract routing in hands, which results from the automated ducting generation. Generally an abstract routing consists a bunch of straight lines as the axis of every straight tube, and they interact with 90-degree corners. Like the following graphic, the orange point usually is located at one of the building chiller plants. The blue points show tranches from this point. The green points provide air for users.

Data Structure of an Abstract Routing

NetworkX is a good choice for someone who’s new to this field. Abstract routing is a path search problem. By the way, if you are interested in Automated Ducting Generation. A* search algorithm and Dijkstra’s algorithm are classic methods to begin with.

Let’s go back to the Data Structure. Recommend reading the NetworkX tutorial. The NetworkX built-in Data Structure graph is a good candidate to save the HVAC abstract ducting. You can customize the attributes of edge and node. For instance, location can be added in node, and weight can be added in node as Euclid distance. By the way, location can be 2D or 3D coordinates. This step is to program the above graphics, which is the foundation of the next Hydraulic Calculation.

Hydraulic Calculation

1. sizing (Duct): one of the goals is to determine the size and shape of every duct. You need to consult the relevant national construction codes (RNCCs), whether the shape is round or rectangular. This step involves data in BIM.

2. sizing (Duct Connectors & Transitions): You must notice the blue points in the above graphics. RNCCs will tell you how to size.

3. sizing (Duct Reducers & Boots): We need to convert abstract ducting into physical air distribution systems, so every edge has a certain amount of air volume. RNCCs require the upper and lower limit of air velocity, which means too long edge might be unable to reach the mandatory, and it should be cut off into two edges.

4. sizing (Valves): The abstract ducting might not consider the valve installation. For sake of hydronic balance, some valves are necessary. In air distribution systems, edges will be classified into the main duct, branch, etc, which is based on the location. You can go through RNCCs. If it is necessary, you can add the categories to attributes of edge, which helps you to add valves. Theoretically, one edge is cut off into two edges, and the new node represents.

5. check (hydronic balance): Our ultimate goal is to ensure that the practical air volume is equal to the theoretical air volume. According to the shortest path to every leaf node, you can calculate the resistance of every patch to check (networkx.shortest_path_length is really helpful). The feedback can guide you on how to modify sizing. Under some circumstances, the abstract ducting needs to be improved, for example, the size of the Duct Accessary (Reducers, Transits, Reducers & Boots) will squeeze the space of the duct. Sometimes the iteration between Hydraulic Calculation and Automated Ducting Generation is necessary.

Practical Tips

I will show you some code blocks to start quickly. Presume you grab a HVAC abstract ducting in hand. First check all nodes and edges in your graph. As a side note, the directed graph would be better than the undirected graph for description.

import networkx as nx
from networkx.algorithms.dag import dag_longest_path_length
from networkx.classes.function import all_neighbors
for edge in graph.edges:
    print(graph.edges[edge].keys())
for node in graph.nodes:
    print(list(graph.successors(node))) 
for node in graph.nodes:
    print(list(graph.predecessors(node)))
for edge in graph.edges:
    print(graph.edges[edge])
for edge in graph.edges:
    print(graph.edges[edge]['airflow']) # if 'airflow' exists
for edge in graph.edges:
    print(graph.edges[edge]['size']) # if 'size' exists
for edge in graph.edges:
    print(graph.edges[edge]['weight']) # if 'weight' exists

About how to add Duct Reducers, Boots, and Valves, you need to locate the new node that means what you wanna add, and replace the inital edge with two new edges. Don’t forget to clone and correct the attributes of edges and nodes .

# Take adding a Duct Reducer as an example
# node_center, node_after are Python tuple
# new_reducing_node is Python tuple
# the next step is to correct attributes, you can customize it
attri_center_dict = copy.deepcopy(graph.edges[node_before,node_center])
attri_after_dict = copy.deepcopy(graph.edges[node_center,node_after])
attri_new_node = copy.deepcopy(graph.nodes[node_center])
attri_new_node['pos'] = new_reducing_node
attri_new_node['airflow'] = 0
attri_new_node['size'] = ()
attri_new_node['type'] = 'reducer'
# cut off the inital edge into two new edges
graph.remove_edge(node_center,node_after)
graph.add_node(new_reducing_node,**attri_new_node)
graph.add_edge(node_center,new_reducing_node,**attri_center_dict)
graph.add_edge(new_reducing_node,node_after,**attri_after_dict)

Conclusion

Acknowledge

1. This work is one part of a larger project about Auto-Design for HVAC System of Public Buildings. The inital floorplan here is one of the test models for this lager project.

2. The first picture cited at the beginning is from Unsplash.com and Photographer Sigmund

Reference

1. GitHub - Autodesk/revit-ifc: IFC for Revit and Navisworks (2019+)

2. Homepage | gbXML

3. Shilin Wu, Yan Wang, Huayu Yang, Pingfeng Wang. (2022) Improved Faster R-CNN for the Detection Method of Industrial Control Logic Graph Recognition. Frontiers in Bioengineering and Biotechnology 10.

4. NetworkX tutorial & documentation

5. A* search algorithm - Wikipedia

6. Dijkstra’s algorithm - Wikipedia