AI将如何进行管道检查？一般而言，AI将提供自动识别CCTV检查中的缺陷并为管道的状况生成准备的检测报告。它的支持者说，人工智能有可能比人眼现在更快，更准确地产生这些数据。
The use of artificial intelligence (AI) that once seemed so far off in the future has quickly and seamlessly integrated itself into the fabric of our everyday lives — to point where we can’t remember not having it at our fingertips. Spam filters. Smart personal assistants such as Siri and Alexa. Autonomously-powered, self-driven vehicles that have predictive capabilities.
It’s only a matter of time before the use AI becomes a fundamental part of the pipe inspection sector. Depending who you talk to, AI technology is either super close or years away from being a part of the pipe inspection landscape. Pipe inspection software companies have been in the research and development stages for some time, working to perfect its inclusion in the inspection process to quickly, and more importantly they say, accurately assess and code the condition of our underground network of water and sewer pipes. Millions of dollars’ worth of decisions concerning pipe maintenance and/or rehabilitation hinge on the data produced via pipe inspections. What if the AI-produced data was incorrect?
RELATED: Water Leak Targeting Using Artificial Intelligence & Machine Learning
While AI’s introduction remains TBD, the thought of how this innovative technology will impact pipe inspection generates talk today. Many believe AI is coming sooner, rather than later and that its impact will be immense.
“Artificial intelligence has the chance to be a game-changer in our industry,” says WinCan general manager Mike Russin. “We just have to make sure this technology is tested properly and not released too soon — destroying its credibility. We can achieve great things with AI and allow for communities to be fiscally responsible with their decisions from the data that is gathered from AI.”
RapidView LLC CEO and founder Rex Robison sees the inspection software market moving closer to developing AI, sooner rather than later. “You’re going to see a shift in that market where you begin seeing automated classification of pipeline defects [and the like] that doesn’t require a user to sit in front of a monitor for hours. That’s where this is going,” he says.
What is AI?
What exactly will AI do for pipe inspection? In general terms, AI will offer the ability to auto-recognize defects from CCTV inspections and produce an accurate grade for the condition of said pipe. Its proponents say AI has the potential to produce this data more quickly and accurately than the human eye does right now. Sounds awesome, right?
AI will definitely take pipe inspection to the next level. But not everybody is ready to roll out AI as the next big thing just yet, saying that there’s more to just having the AI technology to ID defects in a pipe.
“AI should address the purpose issues of efficiency and accuracy,” says POSM Software CEO Phil Cannon. “Efficiency dealing with the time it takes to recognize and code and accuracy as it relates to calculation of risk of failure and risk of consequence,” he says. “If we have 1,000 humans entering a code, many different humans will code the same defect in different ways. AI will code the same defect the same way. AI potentially offers better, more accurate numbers for analysis of future failure.”
RELATED: Last Word – Tapping into Millennials at the Construction Jobsite
And part of that efficiency and accuracy is having NASSCO certification of the process. NASSCO’s Pipeline Assessment Certification Program — better known as PACP — has been an integral part of the pipe inspection process for more than 15 years. Using its 1-5 coding system (with 5 being the most severe condition), PACP has given a universal voice to identifying and recording pipe defects, deformities, deterioration and other issues. Any use of AI technology will need to be able to replicate this very specific coding system.
“AI has the potential to bring greater speed and accuracy to the coding and review of CCTV data,” Russin says. “With AI, we could assist the operators in the field by allowing software to help them make more accurate observations during the inspection process. AI could also provide assistance in the QA/QC process by quickly reviewing hours’ worth of videos in minutes, all while generating accurate observations and condition grades needed for sound infrastructure decisions.”
Human vs. Computer
One of the biggest changes for the pipe inspection market will be convincing end users to trust the results of AI technology vs. the human eye. “AI will affect the pipe inspection industry, as I see it, in only one way. There will not, at some point, be a human being making the defect observation and coding the defect,” says Cannon. “The robot will still go into the camera, the video will be captured [but] rather than a human taking that first shot at observation and coding, that first shot will be done by an AI program at a time other than when the live inspection is done.
And Cannon isn’t saying that is a bad change. “One of the benefits is reduction of human subjectiveness from the defect identifying and coding process,” he says. “AI will make the identification and coding of defects much more consistent. AI will simply code the same defect the same way over and over.”
But end users will have to trust those results, he says.
Wave of the Future
Everyone wants the technology and procedures to be better, faster and more accurate. That’s the mantra you hear from pipe inspection gurus when talking about any inspection software innovations. But the use of AI isn’t as simple as just developing a software program and putting it into service. There are many X factors critical to its success. Beyond NASSCO certification/compatibility, considerations include cost to develop, as well as the affordability to the end user to train and invest in AI technology. Also critical is for AI developers to bank a significant volume of successful AI inspections and analysis to bolster end-user acceptance.
RELATED: Advanced Condition Assessment for Small Diameter Asbestos Cement Pipes
“If we rush [AI] into our market without the proper testing, the results of AI will be poor and will fail to gain market acceptance,” Russin says. “[Right now] AI is still in the training process of image recognition. To achieve the highest levels of accuracy, we must train AI on an abundance of data, which can be a very time-consuming process. Without it, AI will not be as accurate as the human eye is today.”
CTSpec senior sales account manager Armand Hudon concurs with Russin about having an accumulation of successful inspections to show customers. He also notes that “The investment into a new technology [such as AI], the people and the training will be significant.”
Hudon further adds that industry acceptance of such a new and dynamic technology will also be a hurdle that AI will have to overcome. There is usually resistance to change by the players in the market — players that include municipalities, contractors and engineers — he says. Acceptance won’t happen overnight, he adds.
And he’s not alone. Speaking to what could hinder AI acceptance, Cannon says, one word: trust. “Accuracy of the process and trusting the process are two different things. AI will not be fully trusted as 100 percent accurate for years,” he says. “It may take five to 10 years for acceptance … Trusting a computer to properly recognize and code a defect will take time. This data is used in very critical decision-making processes. Do you spend $5 million to dig up this main line on Main Street because an AI program data calculated out to suggest a catastrophic failure? Many people will question that for many years.”
So how far away is the industry from implementing AI technology? No one has a crystal ball but Russin believes AI in some form could be ready in a relatively short time. “I predict we are within a one- to three-year window for complete AI components that can consistently match or exceed human evaluation,” he says.
