Blog: Cloud intelligence on the factory floor

Cloud intelligence on the factory floor

In these years, Amazon, Apple, Google and Facebook are planning data centers all around Scandinavia. Here, the technology giants will process the growing data volumes that citizens and businesses generate on, for example, social media, television streaming and internet searches. But it's not just on the consumer side that many people get used to saving, downloading and running applications across the cloud.
The same trend is seen in the industry, where machines are connected to the Internet, also known as the Industrial IoT. However, when industry machines with data intensive processing are connected to the Internet, there will often be little room for delay caused by data being sent from the machine to the cloud and back.
Applying machine learning in production lines and utilizing intelligent industry vision systems can require hard real-time capabilities and often the processing power of at least a cluster of high-end PCs. This combination pushes the need for applying the super powers of the cloud to the factory floor.
Because, if the vision system of an assembly line or industrial robot reacts only one second too late, it can have catastrophic consequences. The giant data centers built in these years are simply too slow and unsuitable for this kind of data processing.
Instead, a new type of architecture is emerging in these years, called edge computing as on the edge of the machines instead of in cloud - sometimes also referred to as fog computing, i.e. along the ground instead of a cloud in the sky. In edge computing there is still a tight coupling to the cloud, but the edge devices are not dependent on the cloud, they can disconnect from the internet and act autonomously when needed. When they connect, however, they will synchronize data and receive e.g. a better vision algorithm or control loop if the cloud has generated one in the meantime. Since the cloud periodically receives data from all devices it can, based on these data, update the edge algorithms, and there you have it: cloud intelligence directly on IoT devices at the factory floor.

How do we get edge computing in our factory?

A lot of research is going on, e.g. at DTU Compute (Technical University of Denmark), but major industrial players, as Microsoft, are launching real industrial solutions on the market. Microsoft is right at the forefront in this development race and has launched Azure IoT Edge for general availability. Azure IoT Edge is a good example of a fully managed service that delivers cloud intelligence locally by deploying and running machine learning, cloud services and custom logic directly on cross-platform IoT devices that are more off than online.
The best thing is: we have tested it and it works, and it doesn’t have to cost a fortune. We have been able to retrofit the edge computing to existing industry systems such as PLCs and industrial buses (Profibus, CANbus and Modbus). An example of a use-case for a customer:

Today: Company ACME develops and sell production equipment for mass production of cakes. Today the company has no automated quality control of the cakes their equipment produces. The equipment has been sold for 30+ years and is located all around the world.

Future: Retrofit an industrial vision set-up that monitors the baked cakes. This allows for easy wins such as number of cakes produced and corresponding sizes, not to mention a modern cloud interface for the machine or even whole production line. With the Azure IoT Edge architecture implemented, the data from the edge can be sent occasionally to the cloud and in combination with some manual data sets of size, color, quality it enables the cloud analytics to refine its algorithm. When pushed to the edge system it now allows for offline quality assurance on the factory floor.

The future of edge computing

Right now, the concept of bringing cloud intelligence to the factory floor is futuristic for most companies. However, in the coming years the possibilities for much more powerful and intelligent edge computation will be even more extensive. Once again this is highlighted by Microsoft. Their cooperation with Intel in their joint project "Brainwave”, paints the outlines of the future of Industrial IoT. The project focuses on bringing ultra-high speed and parallel computing power to the factory floor by utilizing programmable hardware (FPGAs). This will in the future allow for millisecond crunching of vision-data, digital power conversions from multiple power sources, while having real time machine learning at the factory floor.

More information

Managing Partner, Jakob Appel, jakob.appel@glaze.dk, +45 26 17 18 58

Positioning technologies currently applied across industries:

Global Navigational Satellite System: Outdoor positioning requires line-of-sight to satellites, e.g. GPS: the tracking device calculates its position from 4 satellites’ timing signals then transmits to receiving network
–    via local data network, e.g. wifi, proprietary Wide Area Network
–    via public/global data network, e.g. 3G/4G

Active RFID: A local wireless positioning infrastructure built on premises indoor or outdoor calculates the position based on Time of Flight from emitted signal & ID from the tracking device to at least 3 receivers or when passing through a portal. The network is operating in frequency areas such as 2.4 GHz WiFi, 868 MHz, 3.7 GHz (UWB – Ultra Wide Band), the former integrating with existing data network, the latter promising an impressive 0.3 m accuracy. Tracking devices are battery powered.

Passive RFID: Proximity tracking devices are passive tags detected and identified by a reader within close range. Example: Price tags with built-in RFID will set off an alarm if leaving the store. Numerous proprietary systems are on the market. NFC (Near Field Communications) signifies a system where the reader performs the identification by almost touching the tag.

Beacons: Bluetooth Low Energy (BLE) signals sent from a fixed position to a mobile device, which then roughly calculates its proximity based on the fading of the signal strength. For robotic vacuum cleaners an infrared light beacon can be used to guide the vehicle towards the charging station.

Dead Reckoning: Measure via incremental counting of driving wheels’ rotation and steering wheel’s angle. Small variations in sizes of wheel or slip of the surface may introduce an accumulated error, hence this method is often combined with other systems for obtaining an exact re-positioning reset.

Scan and draw map: Laser beam reflections are measured and used for calculating the perimeter of a room and objects. Used for instance when positioning fork-lifts in storage facilities.

Visual recognition: The most advanced degree of vision is required in fully autonomous vehicles using Laser/Radar (Lidar) for recognition of all kinds of object and obstructions. A much simpler method can be used for calculating a position indoor tracking printed 2D barcodes placed at regular intervals in a matrix across the ceiling. An upwards facing camera identifies each pattern and the skewed projection of the viewed angle.

Inertia: A relative movement detection likewise classical gyroscopes in aircrafts now miniaturised to be contained on a chip. From a known starting position and velocity this method measures acceleration as well as rotation in all 3 dimensions which describes any change in movement.

Magnetic field: a digital compass (on chip) can identify the orientation provided no other magnetic signals are causing distortion.

Mix and Improve: Multiple of the listed technologies supplement each other, well-proven or novel, each contributing to precision and robustness of the system. Set a fixpoint via portals or a visual reference to reset dead reckoning & relative movement; supplement satellite signal with known fixpoint: “real time kinematics” refines GPS accuracy to mere centimetres; combine Dead Reckoning and visual recognition of 2D barcodes in the ceiling.

LoRaWAN: A low power wide area network with wide reach. An open standard that runs at unlicensed frequencies, where you establish a network with gateways.

Sigfox: A low power wide area network reminiscent of LoRa. Offered in Denmark by IoT Danmark, which operates the nationwide network that integrates seamlessly to other national Sigfox networks in the world.

NFC: Used especially for wireless cash payments.

Zigbee: Used especially for home automation in smart homes, for example. lighting control.

NB-IoT: Telecommunications companies’ IoT standard. A low-frequency version of the LTE network.

2-3-4G Network: Millions of devices are connected to a small SIM card, which runs primarily over 2G, but also 3G and 4G.

Wifi: The most established standard, especially used for short-range networks, for example. in production facilities.

CATM1: A low power wide area network, especially used in the United States.

Glaze IoT Cloud Project Process

Beacon Tower is Glaze’s Industrial IoT Cloud Platform that can act as either a stepping stone (Platform-as-a-Service, PaaS) or as a out-of-the-box solution (Software-as-a-Solution, SaaS) for collection of IoT-data.

Beacon Tower resides in Microsoft Azure and is designed as a customisable and cost-effective IIoT Cloud Platform that helps simplify deploying, managing, operating, and capturing insights from internet-of things (IoT)-enabled devices. Our customers have the full ownership of their data.

When running it as a PaaS we utilise the design and can run it on our customers’ Azure tenant and customise it fully to their requirements.

Beacon Tower connects to all sensors, PLC, DCS, SCADA, ERP, Historians and MES to gain maximum automation flexibility and ​prevent vendor lock-in.

For more information visit www.beacontower.io or read the PDF.

Edge Computing Categories and Questions

Device:
o    Sensors
o    Internet connectivity
o    Battery consumption
o    Field Gateway
o    Communication protocols (HTTP, AMQP, MQTT, Gateway)
o    Format of the telegrams sent to the cloud (JSON, Avro, etc.)

Data:
o    Number of devices & number of signals
o    Amount of data to transfer per day
– Event-based or batched or mix
– Transfer rate (every second, minute, hour)
o    Device timestamps
– Synchronized timestamps with cloud or not
– Local buffering on device, late and/or repeated data
o    Any time-critical notifications / alarms
– Latency expectations for non-time critical data
– Alarms generated by device and/or by cloud platform
o    Cloud-to-device messages & commands
o    Analytics
– Results from time-series data / Streaming analytics
– Analytics workflows on data, machine learning etc.
– Edge analytics / intelligence

Cost expectations:
o    Retention periods (for reporting purposes)
o    Aggregation of data, possibilities for cost saving

External integrations:
o    Reference data / online data

Administration, rights and access:
o    Requirements for multi-tenancy (segregated owners)
o    Owners/tenants and operators/technicians
o    Administrating access to data, auditing use
o    API management, consumption of data, 3rd party integrators

Operation:
o    KPI measurements for device
o    KPI measurements for cloud platform
o    Requirements on operators and SLA’s

User-interfaces and functions:
o    Operators/technicians
o    Customers/end-users

Glaze Business Innovation and Development Framework (BIDF)

1. Strategy

Creating an IoT Strategy that aligns with the existing company strategy and/or points out any discrepancies that needs to be addressed. The IoT Strategy should pinpoint type of IoT opportunities that should be sought and how they can support the Company delivering on their overall strategies.

2. Ideation

The Ideation phase is an innovative and creative phase where we identify the IoT opportunities within the company. This is done by using existing assets, industry expertise, industry analysis, strategy and IoT expertise to find opportunities for IoT endeavors. This is done in an structured but open-minded and creative setting.

3. Refinement

In Refinement the opportunities are detailed, prioritized and evaluated in a series of steps with the goal of finding a short list of initiatives the company want to pursue. These steps takes strategy, competence, risk level, customer maturity etc into account during prioritization.

4. Valuation

The short list of opportunities are detailed even further and business cases are created for each of them. This will lead to a decision which opportunity to pursue further.

Moving on from the Business Innovation phases to Development activities we focus on taking the minimum possible risk of building the wrong solution by using agile development practices.

5. Exploration

Proof of Concepts carried out in this phase in order to map out technology as well as user-oriented risks. This also refines the budget and thus valuation and business case. Also giving valuable input to baseline system architecture and eco system involvement.

6. Planning

Moving to Planning phase, the most promising business case has been selected and now it is time to plan the Minimal Viable Product (MVP), in terms of timeline, resources and detailed design.

7. Foundation

Implementing the baseline architecture, toolchains and most critical points of the project.

8. Development

Full MVP is developed using these three principles: Start small, don’t over-engineer; Agile software development – late changes welcomed; Continuous delivery – every change is immediately visible.

9. Operations

Operations in an IoT-project is more than just keeping the product alive. It is life-long updates and continous sharpening of features and business model, meaning new ideas are fed back in the Innovation and Development Framework.

Heat map example on a typical business case: