NDT is a category of analysis techniques used in science and industry to evaluate the properties of a material, component or system without causing damage. Because it does not permanently alter the object under analysis, it is a highly valuable technique that saves time and money in product evaluation, troubleshooting, and research.
NDT is a relatively new form of analysis, and offers significant advantages over its predecessor, destructive testing, which has been the cornerstone of materials testing for decades. Destructive testing involves pushing materials to failure in order to understand performance or behaviour under different loads, and until recently was most suitable for mass-produced products, where the cost of destroying a small number of specimens is negligible. Today, modern NDT is increasingly being used in manufacturing, fabrication and in-service inspections to ensure product integrity and reliability, to control manufacturing processes, lower production costs and to maintain a uniform quality level. In the construction industry, NDT is used to ensure the quality of materials and joining processes during the fabrication and erection phases, and in-service NDT inspections are used to ensure that the products in use continue to have the integrity necessary to ensure their usefulness and the safety of the public.
Currently, there is no shortage of available NDT techniques and approaches. However, techniques capable of delivering rapid, contactless, non-invasive, and high-resolution imaging of subsurface features at a level of only a few microns are still scarce. It is in this space that Sightline specializes.
Based on laser technology, optical coherence tomography (OCT) is a non-destructive, non-invasive, and contactless high-resolution imaging method that allows the acquisition of 1D, 2D or 3D image data of sub-surface regions, with depth resolution of one micrometer or better. OCT can be described as ‘optical ultrasound’, imaging reflections from within materials to provide cross-sectional images. OCT is firmly established in the medical community, particularly ophthalmology, because it provides tissue morphology imagery at much higher resolution than other systems such as MRI or ultrasound. However, its applications extend far beyond medicine, and it is increasingly being used for industrial purposes, particularly quality assurance. OCT’s ability to penetrate most materials makes it a potentially revolutionary technology for industrial quality inspection.
An OCT scan of an art object
The key benefits of OCT are:
OCT’s advantage is that it is based on light, rather than sound or radio frequency. An optical beam is directed at the material to be analyzed, and a small portion of this light that reflects from sub-surface features is collected. Most of this light reflects off at large angles, which distorts the results in conventional imaging. However, using a technique called interferometry, OCT is able to record the optical path length of photons, allowing the rejection of photons that scatter multiple times before detection. This way, OCT can build up clear 3D images of thick samples by rejecting background signal while collecting light directly reflected from surfaces of interest.
Hyperspectral imaging (HSI) is a process that analyzes energy from across the electromagnetic spectrum, with applications in astronomy, agriculture, biomedical imaging, geosciences, physics, and surveillance.
Whereas the human eye is restricted to seeing color in mostly three bands (red, green, and blue), spectral imaging divides the spectrum into many more bands. HSI divides images into bands beyond visible light, where the recorded spectra have fine wavelength resolution and cover a wide range of wavelengths. Certain objects leave unique 'fingerprints' in the electromagnetic spectrum, referred to as spectral signatures. These 'fingerprints' enable identification of the materials that make up a scanned object.
Recent advances in sensor design and processing speed has cleared the path for a wide range of applications employing HSI, ranging from satellite based/airborne remote sensing and military target detection to industrial quality control and lab applications in medicine and biophysics. Due to the rich information content in hyperspectral images, they are uniquely well suited for automated image processing, whether it is for online industrial monitoring or for remote sensing.
A HSI camera operates by imaging a scene onto a slit which only passes light from a narrow line in the scene. After collimation, a dispersive element (for example a transmission grating) separates the different wavelengths, and the light is then focused onto a detector array. The net effect of the optics is that for each pixel interval along the line defined by the slit, a corresponding spectrum is projected on a column of detectors on the array. The data read out from the array thus contains a slice of a hyperspectral image, with spectral information in one direction and spatial (image) information in the other. By scanning over the scene, the camera collects slices from adjacent lines, forming a hyperspectral image or "cube", with two spatial dimensions and one spectral dimension. Note that the scanning is often intrinsic to the application: In remote sensing, scanning is provided by aircraft or satellite movement. Also, in many industrial quality control applications, products conveniently pass the sensor on their conveyor belt.
HSI has been used extensively in the analysis of art objects, as the technology is capable of determining measurements such as age, authenticity, and original color values for restoration purposes.
A hyperspectral image taken from a satellite
A subset of AI, machine learning is a program that extracts specific features from data to solve predictive problems. When powered by machine learning, a computer program can teach itself to grow and change when exposed to new data.
The key aspect to machine learning is its iterative nature. As models are exposed to new data, they are able to adapt on their own, learning from previous computations to produce reliable, repeatable decisions and results. In essence, the machine becomes smarter, faster and more intuitive each time it performs a given task.
You may be surprised at how much of your daily life is affected by machine learning. While examples such as Apple’s Siri and self driving cars are obvious applications, the same technology drives Google’s intuitive search queries, your smartphone’s autocorrect function and even the movie recommendations you get from Netflix.
Other examples include:
Sightline’s SightScope sensor begins with a laser source, which emits a beam of photons into a fiber optic cable. This beam is sent to a beam splitter which, as the name implies, divides the beam in two, sending 90 per cent of the photons to the sample arm (which will scan the surface in question) and 10 per cent to the reference arm.
The purpose of the reference arm is to provide a reference sample of clean, unchanged light that will be compared with the light from the sample arm. Light travels along the reference arm via a single fiber optic cable to a mirror. The light is reflected by the mirror, back into the cable and into a second beam splitter, which will combine the beam with the light from the sample arm.
The sample arm directs the light into a collimating lens, which parallelizes the beam of photons. This parallel beam passes through a focusing (converging) lens, which is pointed at a galvanometer. The galvanometer oscillates within a fixed range, steering the beam along the width of the scan.
The beam bounces off the surface being inspected and is sent back along the same path to the second beam splitter, where it is combined with the light from the reference arm. The light enters the detector, where the divergence in echo times between the two beams displays interference, or what has been detected in the sample. The energy on the detector is then processed, delivering a single A scan. SightScope processes 30,000 A scans per second, yielding several hundred B scans.
Sightline is a leader in AI, particularly the sub-fields of machine learning and deep learning. Both of these technologies require vast amounts of data in order to fully deliver the benefits of AI, and industrial-scale use of OCT and HSI provide ample datasets with which to work. Using non-AI software, a modern factory employing OCT and/or HSI in its quality inspection process would not be capable of leveraging the value of its data. However, a facility equipped with machine or deep learning software could automated image analysis millions of images a day, identify known and new defects, perform upstream analysis to find the root cause of issues and eliminate conditions that lead to substandard products.
It is only through the integration with AI -- linking cutting-edge sensors with machine or deep learning software -- that true Industry 4.0 capability can be realized.
Sightline alone is pushing forward with the integration of AI, OCT and HSI, and we are confident that for enterprises searching for a new standard in industrial quality inspection, seeing is believing.