Capturing and comparing 3-dimensional shape for search engine and other purposes has been attempted by many large and sophisticated organizations ... but without success. CADSEEK is different. It is a highly optimized technology that far exceeds every competitive approach in accuracy but without sacrificing speed, scalability, or security.
CADSEEK was created from the vision of Dr. Abir Qamhiyah, along with the assistance of Dr. Donald Flugrad, the National Science Foundation, and also support from several leading industrial companies. The goal was to develop a technology capable of precisely capturing the geometry (shape) of 3-dimensional digital designs to foster part reuse and enable numerous other use cases.
CADSEEK is an algorithm that mathematically transforms complex 3-dimensional geometry (shape) into a purely numerical signature that precisely describes the shape from which the signature was generated.
Essentially, CADSEEK produces a fingerprint for each design, allowing one design to be used as a search target to find all others like it.
A comparison of CADSEEK signatures instantly produces a calculation for the degree of similarity.
The calculation includes all facets of geometry including size, and provides users instant feedback for the relevancy of each match. It also allows search results to be listed in a ranked and meaningful order.
While other methods are focused on attempting to find duplicate designs, CADSEEK's similarity scoring enables many additional use cases which require similarity scoring, as well as shape-based analytics.
There are many flavors of artificial intelligence. CADSEEK is NOT a classic neural net that requires algorithm training and large training datasets.
CADSEEK instead falls into the AI area of computer vision and digital signal processing, incorporating pattern recognition and reliable mathematics to produce precise results for any domain from engineering to chemistry.
CADSEEK is not "taught" to know that a design is a brake rotor, hub cap, hydraulic cylinder or protein molecule. The intelligence for matching is embedded in each signature. In simplistic terms, the more similar two signatures, the more similar the designs from which the signatures were generated. Intelligent search processes enable the signature of the search target to retrieve all similar designs (signatures) in a ranked order of relevancy in real time. Every search is a live search (not recalled results) performed in under two seconds even in multi-million item databases.
Compared to traditional neural net methods, CADSEEK has several advantages.
CADSEEK is far more accurate than any other shape matching application, in part, because it captures every nuance of a design's shape at the resolution of CAD instead of just a single, monolithic parameter value like surface area.
A useful illustration is the comparison of an orange to an apple. Both can have very similar surface area or volume. While other methods would report the orange and the apple as identical or similar, CADSEEK, in contrast, can see the difference in exterior shape, the number, position and size of seeds, the segmentation of the fruit and so on. Even differences in skin texture between two otherwise identical oranges would cause a change in the CADSEEK Signature.
CADSEEK provides three search tiers: Standard, Expanded and Rough Sketch. It is possible that when an orange is used as a search target, that an apple would be reported as similar in an Expanded search, and probable that the apple would be included in a Rough Sketch search.
Rather than just providing a clue that two items might be similar, CADSEEK provides evidence of the degree of similarity, with such high sensitivity that it can even detect small variances in the level of tessellation in two otherwise identical designs. While finding duplicates is useful, there are actually far more use cases for matching similar designs, such as cost estimation, design reuse, supplier section, and testing plan reuse
The magic of CADSEEK is that its fingerprints have the intelligence to enable an instantaneous calculation of the degree of similarity, even in multi-million item databases. This intelligence enables:
The algorithm that creates the CADSEEK signatures is a one-way mathematical transformation that is a form of encryption. Signatures have been proven to be non-invertible to shape, even by iSEEK. Unlike standard encryption, there are no keys for deciphering, making signatures highly secure. While installations and processes are normally contained entirely within an enterprise's firewall, the security of signatures enables cloud deployments if desired. This security feature is a key enabler of iSEEKGlobal, a supply chain portal that allows OEMs the ability to search supplier catalogs.
Every CADSEEK search is performed in real-time instead of being a result recalled from a previously performed comparison. This means that a new design, such as one received from a customer or supplier, can be used as a search target without needing to first add the design to the search database or the PLM.
CADSEEK is fast, both in terms of encoding files to produce the signatures, and also for search times.
CADSEEK signatures are numerical which enables fast seek times, allowing databases to easily scale into the millions of designs. John Deere's testing of the CADSEEK API showed 0.02 second search response across a multi-million item database.
CADSEEK is not a learning algorithm (artificial intelligence) that has to be trained on thousands of similar items. Instead, CADSEEK is able to capture the shape of any 3-dimensional geometry without any customization of the CADSEEK encoding algorithm. The same algorithm used on 3-dimensional molecular models is also used for tractor parts.
CAD Format: CADSEEK can encode all types of native CAD (both models and assemblies) including PTC, Dassault, Siemens, Autodesk and other CAD formats; Neutral and derivative formats such as STEP, IGES or STL; and even 3D laser scans.
Discrete Parts & Assemblies: CADSEEK can encode designs from bolts to satellite assemblies. The algorithm can even encode non-continuous geometry and weldments.
Industries: The origin of a 3-dimensional digital shape is irrelevant. The identical CADSEEK algorithm has been used to encode designs in numerous industries including automotive, heavy equipment, fixtures, appliances, aerospace, chemistry, biochemistry, machine tools, medical devices, electronics, construction, paleontology, body scans and sculpture.
When designs lack attributes they become invisible to search, but since all designs have shape, no design can hide from a CADSEEK search.
Each CADSEEK fingerprint is numerical and a very small fraction of the size of the original native file, providing massive scalability with minimal storage requirements.
The process of encoding a CAD database to create a shape-search index is fully automated with no human requirement for sorting, tagging, or classifying. As a result, deployments can be performed in days rather than months.
Keeping the CADSEEK Search Index synchronized with the PLM is also a fully automated process that is generally performed on a daily basis.
The first step in performing a shape search is selecting a search target ... a design that is used as a seed to find all others like it. CADSEEK applications enable nearly a dozen different ways to select a search target including attribute filtering, file import, visual browsing, import from CAD, import from PLM, extract from assembly, and search from photographs.
CADSEEK search is also flexible because iSEEK provides a suite of applications, where each fits a differing preferred workflow. In addition, the CADSEEK API allows shape-search capability to be embedded in other applications such as dashboards, catalogs and websites.
CADSEEK fingerprints are identical regardless of the environment or scope of the CAD database they were created from. This means that an OEM could search a supplier's search index without ever having access to the supplier's CAD files.
CADSEEK is not embedded within the constraints of a CAD or PLM system, providing it the ability to comingle data from multiple PLMs, CAD vaults or folders; and attributes can be sourced from the PLM, ERP, spreadsheets and delimited text files.
Independence also allows CADSEEK applications to include visualization and design inspection and compare tools conveniently in the search application, which is especially valuable to users without access to CAD.
CADSEEK operates within the specifications of standard computer hardware, and with surprisingly low requirements for RAM and disk space.
