Modal identification and damage detection – A contemporary real-time approach
New readers, welcome! I am Basuraj and I will be sharing my experience of structural health monitoring research in this blog series. In case you are a new member, I would suggest you kindly visit the first part of this blog series here. I hope you enjoy it!
The second part of the EMI-PMC 2021 experience will focus on the numerical and experimental applications of the eigen perturbation analysis (EPA) based algorithms. Multi-sensor, multi-directional, and multi-dependencies-based recursive canonical correlation analysis (RCCA) algorithm was my algorithm of choice for the EMI-PMC 2021. I will make an attempt in explaining the numerical and experimental investigations carried out using this algorithm purely in an informal approach - which is precisely the objective for this blog series!
With contemporary eigen perturbation solutions steadily gaining prominence, RCCA based EPA has shown tremendous capacity to detect damage and identify vibration modes in real-time. Key findings include damage detection capabilities as low as 10% and modal convergence from 20% sample size!
In one of the investigations using first-order eigen perturbation, a simulated 5 degree of freedom linear system was inflicted with a column tendon tear in the 3rd story. Now, as raw acceleration readings are inept at identifying the damaged location, damage-sensitive features applied on the eigenspace provide an indication of the exact instant and location of the damage. As observed from the figure on slide 8, the top plot indicates the instant of damage at 31s, while the figure at the bottom clearly identifies the distortion at the 3rd graph – thereby localizing the damage to the 3rd floor of the system.
To investigate the robustness of eigen perturbation approaches, we now consider a practical system. An engineering consideration for the ASCE-SHM benchmark involves only translational degrees of freedom – which basically makes our lives easier – by not involving rotational modes. Simply put, we have a 12 degree of freedom model which is excited using white noise – thereby emulating an ambient vibration condition. Our job is to identify the structural modes of vibration using the first-order eigen perturbation techniques.
The streaming acceleration response from each degree of freedom is converted into an eigen perturbative framework. Using EPA, the recursively identified vibration modes can be observed above. In order to measure the correctness of the approach, the modal assurance criteria (or MAC) provide a correlation measure with the theoretical modes. An ensemble average indicates a 0.98 value, which is close to 1, where a perfect correlation is indicated. Simply put - if the value is close to 1, you should be happy!
For specific monitoring needs, individual software licenses can be sold against lumpsum payment. License bundles for complete monitoring solutions with instrumentation strategies, data interpretation, and decision scenario alternatives can be offered. The innovation-to-commercialization transition has the potential to witness a whopping 2.4 billion dollars estimate on the total addressable market, while the serviceable addressable market is estimated at 200 million dollars.
The final conclusions summarizing the key findings presented at the conference are:
The prominence of real-time monitoring has heavily relied on traditional approaches. Through this conference, I envisioned a platform to disseminate my research to academicians, practitioners, and researchers on the recent development of EPA-based solutions that can easily replace traditional methods and incorporated into a hardware platform.
The use of RCCA has been demonstrated as a go-to algorithm for both damage detection and modal identification. The plus side? Online fault detection and localization in a single framework that consistently guarantees the identification of structural modes in real-time. This is an aspect that has not been previously explored.
Putting the numbers into perspective - we have a modal convergence rate of 20%. Now this is highly significant considering most of the other algorithms will require that much sample size to either stabilize or the worse case, to train models which ideally consumes more than 50% and a further 20% sample size for testing.
Virtual conferences are a good way to present but a poor way to build networks or forge relationships. It most so happens that an individual or a group tends to interact with the people they're more familiar with. It is now a good time to accept the harsh reality and come to terms with the fact that this practice might continue for quite some time.
I would strongly suggest interested researchers sign up for the EMI-PMC conference that will be held next year (more details here). With some luck, maybe I'll see you there!
I am open to suggestions and ideas for making this blog series more interesting and appealing. I will look forward to hearing your thoughts. Place a comment for this blog or send in a personal message in the chatbox. Make sure to leave your email ID and I will get back to you soon.
Copyright: Basuraj Bhowmik 2021