automation in construction safety and health
TRANSCRIPT
Sharings from some RnD projects
Automation in Construction Safety and Health :
City University of Hong Kong 2018
Dr. Ivan Fung Assistant Professor, Department of Architecture and Civil Engineering
Why need Automation ?
Why need Automation ?TimeCost QualityQuantity………
> OSH: Construction Safety & Health?
New things on Construction Safety…
Automation =Evolution +
Integration +Optimization +
Ubiquitous
RnD Ideas/ Projects
RnD #1 : AI x BIM (CIM) BIM will be extended to Level 4 and to be more Smarter!
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #1 : AI x BIM (CIM)
RnD #2 : Real-time Monitoring for Construction Workers by Far-field Automated Surveillance Network Camera & UAV
WWW.PPTSTORE.NET
YOUR COMPANY
Findings on Literature Review
Principles• Background modeling, track management, object recognition and light trackingDirectly track human and vehicles, or track any moving objects and focus onto itApplications• Not mature in construction industry• Junction Electronic Eyes to monitor the traffic conditions
Far‐field Automated Surveillance Network Camera
Unmanned Aerial Vehicles (UAV)
• Inspection and monitoring• Surveying and mapping• Condition Survey & civil engineering
• Aerial imaging (HR photos & stills)• 3D scanning
WWW.PPTSTORE.NET
YOUR COMPANY
01Findings on Literature Review
• A set of data are collected at first and allow processing and computing
• Allow computer to learn by examples• Allow multiple processing layers to compute classification
directly from images, text or sound• Deep means the no. of hidden layers in neural network• More than 150 hidden layers
Deep Learning Method
WWW.PPTSTORE.NET
YOUR COMPANY
Findings on Literature Review
Deep Learning Method
❖ Mainly process 2D data like images
❖ Directly extracts features from the image
❖ Speed up and simplify R‐CNN
❖ Joint extractor, classifier and regressor to unified framework and form RoIPool
❖ Regions with CNN features❖ Input an image ❖ Extracts 2000 bounding bo
xes❖ Extract the features for eac
h bounding box ❖ Classify the regions
❖ Deep fully convolutional network and a Fast R‐CNN detector
❖ Add Region Proposal Networks (RPNs)
❖ Reach real‐time
CNN Fast R‐CNN
Faster R‐CNN
R‐CNN
WWW.PPTSTORE.NET
YOUR COMPANY
12Preliminary Design System of TRACE
First trial to use UAV (23/8/2017)
• Identify the limitations of UAV in different environment and construction site
• Select appropriate type of UAV for construction site
Second trial to use UAV, with Dr. Ivan Fung and Mr. Peter Lai (20/1/2018)
WWW.PPTSTORE.NET
YOUR COMPANY
WWW.PPTSTORE.NET
YOUR COMPANY
01Preliminary Design System of TRACE
• Identified the technical requirement and resources for TRACE• Worked on the preliminary design and development of application prog
ramme• Experienced on the usage of existing far‐field automated surveillance ne
twork camera with Mr. Peter Lai and Mr. Sam Huang, Hikvision Oversea Pre‐sales Engineer (22/1/2018)
• Identified the technical problems and limitations of existing surveillance network camera
05
TRACE Prototypes• Allow automated real‐time PPE tracking around const
ruction site and safety monitoringEnsure all workers are well‐equipped with suitable PP
E and permit all‐time safety monitoringIdentify worker who is malpractices and safety warnin
g will be given to the safety officers or supervisors through mobile devicesImprove the traditional visual checkingHighly relief the burden of safety officers
WWW.PPTSTORE.NET
YOUR COMPANY
12Basic setup of TRACE
WWW.PPTSTORE.NET
YOUR COMPANY
Steps for the development of TRACE
❖ Collect photos and videos
❖ Different posture and movement of workers
❖ With/without PPE
❖ Processing of data
❖ Designed for Deep Learning Method
❖ Carry Faster R‐CNN training
❖ Collect photos and videos
❖ Different posture and movement of workers
❖ With/without PPE
❖ Alert the safety officers
❖ Allow warning to worker in real‐time
❖ Correctness of output and modification of program
Collect relevant Data
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5
Design Algorithm for the system
Train the system by Deep Learning Method
Design and program a mobile app for notifying the safety officers
Testingand modification
WWW.PPTSTORE.NET
YOUR COMPANY
12TRACE ‐ Proposed applications
❖ Address all potential hazards Bamboo scaffolder does not wear
safety belt when working at height
Site Safety Inspection on PPE
❖ Identify any action or movement of workers prohibited
Worker is holding a cigarette with his finger and place it onto his mouth, the system will be alerted
Site Safety Inspection on Violation of regulations or rules
❖ Construction site condition can be shown on the mobile devices in real‐time
❖ Directly reach the area or worker for improvement
Site Safety Inspection outside the construction site
WWW.PPTSTORE.NET
YOUR COMPANY
12TRACE ‐ Proposed applications
❖ Hikvision DS‐TCG225❖ Identify the model, licenses, color of vehic
les Record the entry of concrete truck Ensure the workability of concrete
Identify and record the vehicles enter to the construction site
❖ Neither recorded workers or visitor❖ minimize unwanted loss of property
❖ Identify the type and application of construction plant using in different zone
1 bulldozing machines in zone 2
Ensure no people or equipment within the safe zone
No people surrounded in a radius of 1m
Ensure the machine operator is authorized
Monitor and record the usage of construction plants or equipment
Prevent unauthorized people from entering the construction site
RnD #3 : Safety Training & Inspection using MR Technology
Reasons why MR Train is impactful
Consultations with Professionals
Selection of Equipment Equipment for projecting training scenario
MR CAVE RoomHead Mounted Displays (HMDs)
Selection of Equipment Senses Controller HMDs can only provide sight and hear sense Functions: add sense of touch to the visual interface Examples: Haptic Gloves Haptic Shoes
Hand Omni (Rice University, 2015) Taclim VR shoes and glov(Cerevo Inc, 2017)
Selection of Safety Training Scenarios
Type of loadshifting Machinery Bulldozer Compactor Loader Excavator Grader Locomotive Dumper Fork Lift - Commonly used in construction sites
- To lift and move material short distance- Cause serious injury and death
Framework of MR Train for Forklift
6. Finish the course
5. Practical Test
4. Safety Procedures (e.g. inspection and operation)
3. Hazards from operating loadshifting machinery
2. Features of the loadshifting machinery
1. Introduction of Relevant Legislation
Development of Scenario Goal: To create a 360 – degree scenario showing all features of forklift
Development of Scenario
Taking photo by Ricoh Theta S
Photo by Ricoh Theta S
Two special functions
Voice Navigation Magnifier
Development of Scenario Prototype of Session 2 – Features of Forklift
Development of Scenario Session 4 – VR Self Learning Purpose: to understand safety inspection before operation Safety operation
Feasibility Study of ImmersiveMixed Reality (MR CAVE) in RemoteConstruction Safety and Risk Assessmentat Distant Work Sites
MR CAVE Prototypes• Communication network system
– MR CAVE system must be to connect with the trolley on site area. The com
munications network can be the Internet or more precisely the WWW.
The End