BeeYard supports data scientists throughout the entire MLOps workflow from data collection and organization, annotation and labeling, to testing and development to get machine learning algorithms into production in a stable, reliable and scalable way.
Without data, you are just another person with an opinion.
W. Edwards Deming
Modules covering the whole machine learning lifecycle
BeeYard consists of many modules that help you at every step of your machine learning project. Development of a robust machine learning model begins with the creation of well-formed dataset provided to all stakeholders via secured & managed access. Very important but also least enjoyable is proper annotation & labeling of the dataset. We save your time and resources with our AI assisted labeling capabilities. Automation pipelines enable you to delegate the repetitive tasks to our robots. You can leverage our hybrid cloud functionality to transfer data between at-the-edge and cloud infrastructures. For large machine learning tasks and testing of different hypothesis you can decide to use our cloud infrastructure or your private cloud.
01. Data Collection
BeeYard provides many connectors out of the box to help you collect information from various sources. The platform originates from machine learning & machine vision background with deployments in the manufacturing environment. Besides connectors to all major databases, like many other platforms, BeeYard has a first-class support for data originating in the manufacturing process. We have a built-in support for many industrial cameras, smart-cameras and PLCs.
BeeYard is ready for legacy devices by providing different gateways (e.g. an FTP gateway, Samba gateway and others) that can be used to collect data.
Last but not least, BeeYard provides SDKs enabling developers to easily extend their solutions with data management & machine learning capabilities. For low-level integration with BeeYard REST API is available.
02. Data Organization
BeeYard provides many capabilities to keep your dataset secure & consistent. You can organize your data into workspaces with pre-defined constraints.
BeeYard is a multi-user system that provides roles-based access. Using our fine-grained permission options, you can precisely define which user or role can access which part of the system. Access can be granted on the level of individual data cells.
Pre-defined, workspace-wide labels and other annotation tools can be specified to keep your data consistent and reduce the error rate of human annotators.
BeeYard provides powerful annotation toolset that dramatically speeds up the time required for annotating a dataset. Annotation jobs can be distributed to annotators anywhere across the globe.
Using the AI assisted labeling capabilities enables human annotators to get their job done 8 times faster than they would achieve using previous techniques.
04. Training & Execution of your models
BeeYard lets you train your deep learning model or execute your deterministic algorithm on your large datasets. By doing so, it automatically collects performance statistics for you.
Thanks to its open architecture, BeeYard is an ideal choice for building composite AI models that consist of non-deterministic algorithms combined with deterministic algorithms. Such models can benefit from improved performance and better explainability.
Each model that has been trained or executed is stored and versioned for future reference.
BeeYard provides large set of automation tools that save you from loosing your time with repetitive tasks. You can set up your automation pipelines that get dynamically triggered by important events. Among the most common automation tasks are:
BeeYard provides you with extensive monitoring capabilities. Keep track of your mission critical deployments by real time monitoring of your models' performance. Be always notified of important events by setting up your business-specific alerts.
Have insights into your datasets by generating on-demand reports and statistics.
Maintain full visibility over operations that have been performed within your account. The audit log lets you see the history of any record ever stored into the BeeYard platform.
BeeYard empowers you to leverage the full power of cloud computing. Developers' desktop is not always sufficient when dealing with large datasets or doing resource-intensive operations like teaching a new deep learning model. BeeYard is optimized for developers' round trips and keeps algorithm development iterations short by simply shifting the workloads into more powerful infrastructures. The development of mission-critical machine learning models is significantly faster using this technique instead of using less powerful in-house infrastructure.
BeeYard is a collaborative platform that allows users in various roles to interact on the same project. It enables use cases like distributed annotation, quality assurance workflows, collaboration of multiple development teams operating on the same ground truth and many more.
Collaborative annotation user experience is boosted by the concept of annotation jobs that let you split the annotation task and distribute it in form of smaller batches. An annotation job must comply to quality assurance criteria defined by the quality manager.
BeeYard delivers measurable business value
BeeYard helped to accelerate the production lines of a major car manufacturer by a contactless robot guidance system.
Product Quality Improvement
BeeYard helped to ensure top product quality by automated package content verification at major furniture manufacturer.
BeeYard helped to digitalize manual internal processes and back-office operations while ensuring significant time and costs savings at fintech industry.
BeeYard helped to reduce the number of customers’ claims by an automated quality inspection system in tire production.
Insights from Data
BeeYard helped to build a robust OCR model for automated reading and verification of various product labels and codes.
BeeYard helped to replace inaccurate manual control with highly precise contactless inspection of pin connectors.
Featured case studies
Contactless optical localization of car bodies on welding lines
BeeYard helped to significantly increase production effectiveness by reducing cycle time and costs associated with vehicle bodies on car production lines.
Static robot guidance is commonly used at production lines when welding car bodies at robot workstation cells. The car body must be mechanically lifted and placed in a precise position. The ultimate position of the car body is pre-set using the RPS (Reference Position System) and cannot be changed dynamically. The car body is positioned by lowering the RPS holes in the car chassis over guiding pins. A mechanical clamping device is used to hold the car body in a precise position. The technology is not only demanding for the initial investment but also prone to mechanical damage and increased maintenance costs. In addition, the whole process is not efficient because the vehicle body must be mechanically lifted, which consumes production time.
BeeYard was used as a platform for development of a robust contact-less solution for dynamic robot guidance utilizing machine-learning and AI capabilities. The ultimate requirement was to optimize the process while reducing production costs as well as the risks of mechanical failure.
Dynamic guidance of welding robots means that every time the car body approaches the robot workstation cell, the robots need to adjust their welding paths to the actual location of the car body. This is enabled by a powerful machine learning model that estimates precisely the current position processing data from 2D and 3D images combined with pre-taught position data.
The role of BeeYard
With fast loops over the data, it was easy to quickly identify the degradation of the algorithm and find the right track to the best performance.
It was crucial to keep track of all the data coming from multiple factories. BeeYard helped with the automation of the data transfer and automatic data sorting.
With one ground truth database, it was possible to work in parallel on the algorithms in a large team and compare the performance of different approaches to choose the best option for the customer.
BeeYard is part of the system CI/CD pipeline and all builds and releases are first tested on the database to ensure that the system performance is getting better.
Powered by BeeYard
Contactless in-line optical inspection of tire marking
BeeYard helped to develop a solution to significantly improve production effectiveness by increasing production quality and decreasing manual work associated with the inspection of tire marking.
Various markers, such as color and adhesive markers, are used as orientation guide while mounting tires on the rim and must therefore be clearly visible and valid. Tire manufacturers must ensure that the pieces produced have markers of the required quality. Otherwise, they risk a complaint, where the customer refuses to take over the entire manufactured batch. So far, the quality inspection has been done manually and has not ensured 100% reliability that the markings on the tires are really free of defects. This generated high additional costs both associated with the manual inspection and with the costs of complaints and returns. Also, the reputation of tire manufacturers and overall brand awareness deteriorated if the quality of supplies did not meet the expectations of customers of these premium brands. The ultimate goal was to automate the manual process of quality verification of each single tire on the manufacturing line utilizing robust machine learning algorithms for overall vision control and monitoring of the production.
BeeYard helped to develop a robust optical solutions for automated quality control and detection of defective pieces. The system leverages the power of a robust machine learning capabilities that deliver reliable results even for demanding industrial conditions such as various illumination, dirt on the tires, or light reflections on the acquired 2D images.
The deployment of an automated vision system makes the process of quality assurance much more reliable and precise as no defective tire goes undetected.
The role of BeeYard
For the successful implementation of reliable AI models, good communication between customer and integrator is mandatory. BeeYard was used to communicate with the customer and resolve uncertainties about production quality.
BeeYard enabled the insights into the production based on acquired data and thus help to improve the production quality.
For the improvements of the tire inspection, it was important to understand the reasons for false detections. With all data stored in BeeYard, including the actual setup of the system, algorithm thresholds, and raw data it is easy to replay the decision sequence of the system and implement corrective measures.
Based on the data feed from the production line it is easy to diagnose the system and solve the incident remotely or better coordinate local maintenance of the system. We were able to solve more than 95% of claims remotely and thus saving costs.
Powered by BeeYard