ArchFX Analytics

Hybrid physics and ML/AI-based models to predict, understand, and improve your operations

ArchFX Advanced Data Analytics and Predictive Tools

Key performance indicators only quantify your current state and identify areas for improvement. Manufacturers want to improve organizational performance and need analytics that identify actionable root causes of diminished productivity.

Until recently, Surface Mount Technology (SMT) manufacturers have not benefited from advanced statistical analysis, including ML/AI, to improve manufacturing operations and maintenance practices.  Historically, this was due to a combination of two factors (1)  the lack of richness in the data available from machines and (2) the difficulty in extracting value from the data once it becomes available.

But all that has changed. The advent of big data analysis techniques with machine learning and artificial intelligence algorithms has largely removed the traditional limitation of being able to extract great value from rich data. And ArchFX, with a library of connectors for rich SMT data, now provides an automated and scalable way to collect and process rich machine data into these ML/AI pipelines.  Manufacturers worldwide are now turning to these techniques and the experts that know how to navigate them to leverage untapped value across their existing factory lines.

From data overload to data driven

Modern, cloud-based data management software is designed to handle petabytes of data without issue.
The data volumes generated inside of EMS factories, even large ones, are easily handled.

SMT assembly lines already generate a massive amount of data from Solder Paste Inspection (SPI), Pick and Place (PNP), Automatic Optical Inspection (AOI), and In-Circuit Test (ICT) machines, among others. Each of these assets and their components have their own real-world duty cycle that is not being analyzed and accounted for today. Traditionally, the large volume of data has made it hard to determine which data matters and is actionable.

Vendor products often have built-in menus that give access to the data, but operators are busy keeping a line running at full speed. They have no time to painstakingly go through menu after menu to check on the performance of the hundreds of feeders and nozzles that may be present on a single line or to analyze hundreds of product build sessions to detect problems like imbalances in recipe design. Modern analytics are needed to extract the signal from the noise so the signal is understood, prioritized, and becomes actionable.

Feeder and Nozzle Analytics

Arch feeder and nozzle analytics rapidly detect mispick rates for specific feeders or nozzles as they exceed acceptable limits, and use machine learning to predict upcoming failures before they occur.

PnP machines use feeders to feed components on reels into the PnP for vacuum nozzles to pick from the reel and place on the board. If a feeder or nozzle begins to malfunction, it can cause failed picks (mispicks) that cause an ever-larger percentage of components to be dropped and scrapped, which is referred to as attrition. Worse, if the nozzle is malfunctioning, even if it doesn’t drop the chip, it may subtly mis-position a chip in ways that may cause the product to fail AOI inspection or even to fail early in the field.

To keep the line running smoothly, PnP machines typically respond to a mispick by repeating the pick attempt. This prevents the line from stopping and helps ensure that throughput and Line Utilization targets are met, but it wastes money on attrition and hides the underlying problems. Pick and Place machines often have thresholds set so that if the mispick rate for a particular feeder rises too high, an alarm may be raised. But busy operators under pressure to meet delivery deadlines and Line Utilization targets often just override the alarm. Line managers and maintenance engineers are unaware that a problem even exists, and problems may persist for days or weeks on end, wasting valuable material and reducing operator productivity by wasting their time on repeated overrides.

Session Analytics to optimize the product build process

Session analytics divide up the day’s Pick and Place activity into “sessions” of continuous, uninterrupted build activity.

Grouping the build activity by continuous build sessions in this way helps determine how efficiently the line can build  product when it’s not being interrupted by errors and other problems. It then becomes possible to determine how much each separate source of loss is reducing line utilization.

Categorizing the sources of line utilization loss in this way enables staff to prioritize and tackle the sources of loss to increase utilization as quickly as possible.

Predictive Maintenance and Predictive Quality

Rich historical data sets to develop models for predictive
maintenance and deeper predictive analytics

Once SMT manufacturers have a clear picture of their manufacturing operations at every level and operators are rapidly recognizing and resolving problems when they occur, the next step is to prevent problems before they occur through predictive analytics. Fortunately, Pick and Place machines provide a rich data stream about their performance and the results of every placement attempt. For every placement error, the program, feeder, slot or track, table or stage, head, segment and nozzle is recorded, and there are dozens of error codes to categorize the errors by type.

Recording just one day’s production for an SMT manufacturer will record millions of device health data points in the data lake, so recording a month or more of activity provides a rich data set to develop predictive models that can predict when PnP components like nozzles and feeders are about to fail and need service and may predict the quality of finished products as well.

ArchFX is the data engine that drives hybrid
physics + ML based maintenance analytics

ArchFX is the bridge from legacy machines and computing
to the promise of Industry 4.0

Automated data collection from machines

Arch provides integrations to smart and legacy machines (including where new hardware is needed) to source the right data for predictive maintenance.

Machine model to identify equipment insights

Arch combines sensor data and context data into a hierarchical model which allows data to be stored and insights to be generated specific to each machine and its parts.

Usage and condition based maintenance

Set triggers based on usage or on conditions in the data. Arch data science identifies latent conditions that are not obvious to the human eye.

Long-advance predictions are actionable

The hybrid physics and ML-based approach makes long-advance, 30 day+ predictions possible. With sufficient prior notice, manufacturers can schedule maintenance to achieve significant ROI.

Yellow warning, red alarm for maintenance

Generate two simple warning signals as the outputs of a complex process.  A yellow warning gives a 30 day+ warning that a machine or equipment is likely to have a failure.  This allows the manufacturer to schedule work and prepare accordingly.  The red alarm indicates high risk of failure now so rapid action can be taken.

Big data advantage and remote expert assist

By liberating trapped data into the cloud, data from across all your sites helps improve your optimization at each individual site. The more machines, lines, and sites you have, the more likely you are to have good predictions. Moreover, a single group of remote experts can be empowered anywhere in the world to review issues and actions at any site. This allows you to centralize asset management and expertise gaining further benefits on implementation costs.

Analytics as a service by Arch

Arch’s Advanced Data Analytics team partners with our customers to understand your data and use cutting-edge machine learning and artificial intelligence techniques to implement Industry 4.0 use cases, such as defect prediction, predictive maintenance, and data-directed automation.

By using an Arch system to connect legacy and new machines, novel actionable insights become possible that weren’t available to your organization before. For example, implementing Arch technology to extract data from both processing machines and testing machines can enable the classification of front-of-line condition monitoring data with end-of-line test data. By building a defect prediction classifier, we may be able to produce massive savings together, identifying defects early and acting to correct them before expensive scrap is created or product destined to fail early is shipped to customers. Reach out to us to discuss your analytics challenges.

Open APIs to develop your own analytics

ArchFX Cloud has RESTful APIs for accessing the normalized data stored in relational databases, time series databases, and the ArchFX data lake. Because ArchFX Broker has normalized the data and ArchFX Cloud has securely stored it for high-performance access, your data science team can focus on semantics and inference instead of data cleaning. Our SDK supports interfacing with modern Python-based data science tool stacks and Jupyter notebook-based environments. Start a trial and see what makes our advanced factory data analytics the best fit for your business.

Developers wanting to use Arch’s rugged, low-cost IOTile PODs without the ArchFX Cloud factory model can order IOTile PODs through our IOTile Developer Store and use them with IOTile Cloud, our standalone IOTile developer cloud. Check out IOTile Analytics on Github and read the documentation on how to do your own advanced analytics with IOTile technology.

Analytics via third-party tools

 

 

ArchFX Cloud’s standard and secure REST APIs make it easy to integrate third-party dashboarding and advanced data analytics predictive tools like Tableau, Thoughtspot, AWS SageMaker, Azure AI, Google TensorFlow, and any other third-party tool of your choice.

Reach out to our team to start collecting and analyzing data today based on proven use cases or to discuss the use case you want to develop. .

Your partner for digital transformation and operational excellence

Arch Systems is your partner to implement affordable, standardized data collection for machine cycles, downtime reasons, and performance anomalies, building up an easy-tounderstand and actionable record of machine lifetime that can be translated into practical predictive maintenance. The end result can be over 10% savings on yearly CAPEX, which for many cost-conscious manufacturers is transformational.

ARE YOU READY TO SEE
THE VALUE THAT’S TRAPPED
ON YOUR SHOP FLOOR?

Find out how you can leverage your own data now.

THE LATEST FROM OUR TEAM

Finding Hidden Utilization in Manufacturing Machine Data

  The Cost of Lost  Asset Utilization Lost utilization is expensive. It’s as simple as that. Every misplaced part, every recipe imbalance, even if imperceptible, costs money. This is especially true for contract manufacturers with electronics manufacturing...

2020: A Year of Growth and Connection

The accomplishments of our team this past year represent something more than hard work by exceptional people. They represent the unification of a diverse, distributed team working through extraordinary circumstances and times, in an industry marked by an indomitable spirit, supported by world-class partners and customers.

The Hidden Cost of Data Silos

In many cases, the limiting factor for how quickly you can solve problems with data is how quickly you can put together a clean, unified dataset. Once you have clean data, it’s straightforward to get value from it.