Figure 1. Screenshot of a channel CCTV video using VAPAR's auto-encoding system © VAPAR.
Inspections and condition assessments of network infrastructure are critical for water utilities and municipalities to ensure the structural integrity and functionality of sewer and storm sewer pipes. These assessments are necessary to identify pipes that require rehabilitation before their condition deteriorates to the point where replacement is no longer necessary.
As a result, water utilities and councils across Australia spend millions of dollars each year maintaining their wastewater and stormwater assets. The traditional CCTV assessment method for network assets presents challenges for utilities and operators, such as:
- the time required for visual inspection of CCTV inspection videos and Identify defects.
- the experience and subjectivity of the operator.
- tracking, locating and organizing large amounts of CCTV data.
In this context, VAPAR CEO Amanda Siqueira and CTO Michelle Aguilar founded the company in 2018 to provide intelligent solutions that automate the condition assessment of network assets. Ms. Siqueira and Ms. Aguilar developed an algorithm that can automatically encode CCTV videos using artificial intelligence (Figure 1).
“The technology offers an accurate, cost-effective and fast alternative to the time-consuming video inspections I worked with during my time as an intern,” says Ms. Siqueira.
As a long-standing provider of network services to municipal water utilities, Veolia is always looking for innovative digital technologies that can improve service quality while reducing labor-intensive tasks. A partnership with VAPAR was therefore the logical step to address the challenges that operators face on a daily basis.
“Veolia is fully committed to digital transformation and is always looking for new ways to use data to improve results for both sides. our customers and Veolia’s operations,” says Veolia General Manager Water Victoria Jean-Michel Seillier.
This prompted Veolia to conduct a trial in Victoria in February 2020 to evaluate the accuracy of the VAPAR software and provide a market-leading case study for the application of AI technology in pipeline inspections.
The positive results of the trial point to exciting developments for Veolia, VAPAR and Australian water utilities and councils.
The process
This trial was based on 198 videos representing 3.6 km of pipes selected from various Veolia Network Services (VNS) inspection campaigns in Victoria, covering a wide range of diameters and materials. The condition of these pipes was assessed by operators during these campaigns.
In parallel, VAPAR used its algorithm to detect defects in all the selected videos. Two assessments were then available for the same video: the “Operator Assessment” and the “VAPAR Analysis”. Veolia then commissioned an independent experienced expert to carefully review these 198 videos and provide a “reference” that served as a source of truth.
The operator rating and VAPAR analysis were then compared with the reference.
To facilitate this comparison, the different defect types were grouped into six categories: cracks, roots, blockages, joint defects, connection defects and others. The performance of the assessments for each of these categories was then estimated.
The following two indicators were calculated to evaluate the performance of both the operator assessment and the VAPAR analysis against the 'reference':
- Precision: Precision is an indicator of the number of “false alarms”. For example, a precision of 80 percent means that 20 percent of the defects listed in an assessment are actually not defects (i.e. false alarms).
- Recall: Recall is an indicator of the number of defects overlooked. For example, a recall of 70 percent means that 30 percent of the actual defects were overlooked during the assessment.
Results
The result of this analysis showed that the VAPAR algorithm performs relatively well, missing only 13 percent of defects compared to 37 percent for the operator. This is mainly due to the ability of the VAPAR algorithm to detect micro-defects that are usually not reported by busy operators.
Figure 2. Recall (missed defects) and precision (false alarms) for each category, comparison between VAPAR and the operator.
The algorithm proved to be particularly powerful in detecting roots, cracks and joint displacements. As shown in Figure 2, the VAPAR algorithm is particularly oversensitive to displaced joints: about 50 percent of the defects reported by VAPAR were not clearly visible in the video images.
Figure 2. Recall (missed defects) and precision (false alarms) for each category, comparison between VAPAR and the operator.
When assessing the condition of network assets, the operational and structural classes of the pipes are particularly important. These classes, ranging from 1 (good condition) to 5 (critical condition), are typically used by water utilities and municipalities to guide maintenance and rehabilitation programs.
The most valuable result of this trial was that the VAPAR algorithm significantly improves the accuracy of the classification. In fact, Figure 3 shows that the VAPAR structural class assessment is correct for 80 percent of the pipes, compared to only 48 percent for the operator assessment.
Figure 3. Overall comparison of VAPAR and the operator regarding defect detection, pipeline structure qualities and service qualities.
Similar results were achieved in the service assessments, with 76 percent of the assessments correct in the VAPAR analysis compared to only 52 percent in the operator assessment. This improved accuracy will enable optimization of network asset maintenance and remediation programs.
In its search for innovative technical solutions that can help improve operational performance, Veolia was delighted to conduct this successful test with VAPAR. The VAPAR algorithm improved the existing methodology for assessing the condition of sewer pipes and stormwater drains.
In particular, the ability to detect micro-defects can significantly refine the health assessment of these critical network resources. Despite a slight oversensitivity, this test demonstrated that the VAPAR algorithm can assess network resources more accurately than operators.
Since structural and service assessments typically serve as a benchmark for guiding maintenance and rehabilitation programs at water utilities and municipalities, this improved assessment can help better optimize investments in this critical public infrastructure.
For this reason, Veolia has partnered with VAPAR and is currently working on a range of offerings to help municipalities and water utilities maximize the return on their investments in wastewater and stormwater facilities.
This article appeared in the September 2020 issue of Trenchless Australasia. Click here to view the magazine on your PC, Mac, tablet or mobile device.
For more information, please visit the Veolia website.
If you have news you would like to feature in Trenchless Australasia, contact Deputy Editor Sophie Venz at [email protected]
Comments are closed.