The BIG problem with BIG data is that the BIGGER it gets the harder it is to see the BIG picture.
This problem is especially true for CAD data. Typically, the only classification scheme for CAD models is text-based attributes, which are rarely complete or uniformly applied. But even if attributes could be complete and uniform, two items labeled as 'valves' can be so different that applying analytics is a waste of time.
But every model has shape...,
... and the more similar any two parts are in shape, the more likely that patterns exist for cost, price, process and performance, which all have strong correlations with shape. CADSEEK's ability to precisely rank models by their geometric similarity allows valuable queries and insights.
What duplicate parts can be can be consolidated ...
Where are the best opportunities to develop a standard design...
How much part duplication exists with this acquired company ...
Should these aluminum parts that are 98% similar have a cost that varies by 37% ...
Who has the most experience designing parts like this ...
CADSEEK's unique ability to PRECISELY score models for similarity allows shape to become a new basis for classifying models, enabling an analytics capability that is no longer compromised by inconsistent and erroneous attributes.
Goals and Features
CADSEEK Analytics report the similarity of each model in the dataset to every other model. Beyond just cleansing data …, the key goals of CAD analytics are using the shape-similarity of parts to spot variances in part metrics, and to identify opportunities for consolidation and standardization…, which lead to both product improvement and cost efficiency. Features include:
The analysis is performed in a completely automated fashion. In a dataset of 100,000 models Analytics will perform 9.99B comparison.
Cross Dataset Analysis
A report can be run within a single dataset, or across datasets. Running reports across datasets allows the comparison of the parts of one division to another, or of the parent company to an acquisition.
User-Defined Similarity Cutoff
Each time a dataset is analyzed a different similarity threshold can be set, e.g., show all models with at least 91% or greater similarity.
Each time a dataset is analyzed a different set of filters can be applied to slice the data with any available meta-data field and attribute value, in combination, and with numerous Boolean operators, allowing a report to be limited to a certain material type, surface finish, author, commit date, division, etc.
CADSEEK Analytics can be performed within or across datasets with a wide mix of different CAD formats, assembly models and neutral formats like .igs, .stl and .stp.
Interactive and Integrated
Any model can be inspected and viewed in 3D from within the report, and links to CADSEEK Connect allow viewing of model attributes and search.
Each Analytics report that's generated can be viewed in three different formats.
The Dataset Overview report shows the level of duplication and similarity within or between datasets. The example report at right shows the dataset has 2,748 pairs of duplicate items, and 229 pairs of items that are from 99% to 99.999% similar.
Dataset Overview Report
The Pair-Models Report shows each pair of models within a dataset or even across datasets, which have similarity above the threshold set for that report
e.g., show me all models with similarity of 85% or greater.
The paired-model format means that if model A1 has three similar models at or above the similarity threshold, then model A1 will appear in the left column of the report three times.
Grouped -Models Report
The Grouped-Models report shows the same data as the paired-models report, but the data is presented on a singled row. For example, if model A1 has similarity to three other models at or above the threshold, then A1 will appear in the left column just one time, and all three similar models will appear on the same row to the right.
Like the Paired-models Report, the grouped models report can be run within a dataset or across two datasets.
Each time an Analytics report is run the user can choose any combination of attribute filters to slice the data and make customized reports, i.e., a report for a division, manufactured parts, a specific vendor's parts, a part with the attribute 'valve', etc.
Each Analytics report can be exported in a tab-delimited format so that the similarity data can be aligned with other data such a s cost or performance, to analyze variances.
The example at right shows a search imported into a spreadsheet, aligned with cost data, and a variance in cost computed.
Other Reports in the CADSEEK Analytics Application
CADSEEK provides the ability to import virtually an unlimited number of meta-data fields and attribute values. The Meta-data Report lists each meta-data field, all attribute values that exist for each meta-data field, and each model that has been tagged with that attribute value.
Because each model is visualized with a thumbnail for each attribute value, it makes it incredibly easy to spot attribute errors, e.g., that's clearly a type F and not a Type K.
The Meta-Data Report creates an incredibly useful road map for data cleansing or data migration projects.
Data migration and data cleansing projects are often bogged down by the need to examine each assembly file to determine what models are required for assemblies. CADSEEK Analytics automates the task, listing each missing model, and which assembly or assemblies it is used by.
Missing Dependencies Report
It's common for a dataset to contain models that lack geometry. Analytics automates this process of identifying these models to reduce clutter in the dataset.
Uncoded Models Report
Use Cases for CADSEEK Analytics
Duplicate parts create unnecessary carrying costs of which can amount to several thousand dollars per year. CADSEEK Analytics reports quickly identify duplicate models providing a road map for investigation.
Identification of Similarity
The Analytics Similarity report provides a listing for each model of every similar model down to a user-defined threshold - e.g., show all models with similarity of 87% or more. This makes pools of similar parts easy to identify and investigate the opportunity for consolidation.
Acquired companies create substantial opportunities to consolidate both manufactured and standard parts. CADSEEK works with any CAD format, so models can be readily compared to find both duplicate and highly similar parts.
PRICE VARIANCE DETECTION
Shape is highly correlated with manufacturing cost or purchase price for highly similar items (with allowances obviously for material and finish). Groups of highly similar models can be exported for alignment with cost data so that cost discrepancies can be easily identified.
DATA CLEAN-UP & MIGRATION
Analytics can be used to clean CAD databases prior to data migration or at any time. The Analytics module provide reports that spot duplication and attribute errors so that such errors can be eliminated. Analytics data can be exported by shape classification so that similar parts can be scripted for the assignment of standard attribute values.
Analytics can examine the work of consultants to check if migrated data still has duplication, or if any data to be discarded is unique and should be kept. For projects to apply attributes, Analytics can determine the similarity by attribute tag to check for outliers.
Return on Investment
Analytics is able to automate work that would take humans thousands and thousands of hours to complete. For example, if a business with 100,000 models is acquiring a business with 50,000 models, an Analytics report to compare the two datasets would be the result of 2.5B comparisons, which might take a couple hours or so to produce. For the human-based approach..., suppose 90% of the 2.5B comparisons could be eliminated because of good quality attributes, and the remaining 250,000,000 comparisons could be completed in just1 minute each.., then the project would require over 2,000 years for a single analyst to complete at a cost of over $130 million.
Projects of such scale would obviously never be funded. But the ability for CADSEEK Analytics to automate the process and create a project road map in just a few hours changes the equation and makes such projects not only feasible but highly profitable. The ROI potential for two common analytics projects are shown below.
1. Eliminating Duplication. The elimination of functionally duplicate parts creates economies of scale and eliminates unnecessary purchasing and inventory carrying costs. The assumptions used in the table at right for redundancy rates and carrying costs are based on research performed by Aberdeen Group and the Parts Standardization an Management Committee of the Department of Defense.
"According to our research, as many as 30% to 40% of manufacturer's parts are duplicates or have acceptable substitutes."
A study by the Parts Standardization and Management Committee of the Department of Defense calculated a five year inventory carrying cost for an average part of $3,750 per part.
2. Vendor Price Discrepancies. While companies routinely purchase duplicate parts from more than one source, the price charged by each supplier should be very similar. When they are not, Analytics makes it easy to group and align these identical or highly similar parts with cost information to spot pricing variances like the one shown in the table at right.