Many manufacturing firms remain unclear about the economic benefits derived through investments in Digital Industrial transformations in the past years. There could be many reasons for that clarity issue. But, they agree and understand that to meaningfully measure an economic benefit through such investments, it will take some time. Like for any industry, Digital Transformation is purely a technology oriented offerings coupled with necessary process change within the factory setup where relevant.
Most of the factory setups implementing Digital Transformation for Smart Factories often does in phases due to budget issues and production priority. The vendor should have good phase wise transformation deployment experience. Transformation solutions vary with vendors because solutions are basically an assembled collection of multiple technology products integrated with each other to derive the benefit and each vendor has a specific technology priority either due to partnership or skills availability. The total price of the transformation is the sum of individual assembled technology product price plus the implementation cost. If one were to choose a vendor who has the skill and the ability to assemble the technology product based on best of open source stack, the transformation cost will be competitive and at the same time neutralize the vendor out of the equation once it is implemented because there are many other vendors available to support open source stack based transformation solutions.

Industrial machines have certain level of capability to support instrumentation systems to measure general operating parameters. The sensor instrumentation systems feed measurements to signal aggregators like PLC within the factory setup.
There is always an implicit relationship between different operational parameters within the machines that either individually or collectively impact the performance. Machine operators use machine manuals as guide to periodically check sensor measurements to ensure the machine parameters are within the range. The measurement systems send sensor readings to the PLC which then aggregate them and display in the panel. PLC systems have narrow processing ability and are not designed to identify co-relation between different sensor readings based on historic events in case of abnormal behavior. For example, sensors are available to measure temperature of lubricants, quantity levels of coolant oils, speed and temperature of a rotational moving part etc, In complex machinery, there are capability to  measure 10 to 15 parameters. Factories are dependent on external technical experts for periodic maintenance tasks. The technical experts are always busy going around factory floors across the country. All types of equipment—machines, batteries, and other mechanical and electrical components degrade with time and use.companies have used schedule-based maintenance, condition-based maintenance methods. Using machine learning, one can monitor multiple parameters and accurately predict when equipment is likely to fail and take preventive actions to save cost and improve productivity.

Implementing Machine Learning and AI in the industrial world requires capabilities in combining domain knowledge in managed factory assets like machinery, lines and plants with probability techniques, machine learning methods and deep learning frameworks to operate on both limited and uncertain data or massively streamed data. Primary objective in such initiative after completing assessment is to design and create a digital twin of the existing industrial asset. A digital twin is a virtual representation of all the managed factory assets. Our capability to use technology and skills to create a digital twin of an industrial asset using pre-build library of DNN models and in which each model will be customized to represent every single managed asset in the factory.
Once managed assets within a factory setup are shortlisted for modeling, we use either hybrid modeling or data driven modeling techniques. Hybrid modeling techniques use combination of physical equations and probabilistic ML techniques. Pure data driven techniques use the advancement of probabilistic techiques in Deep Learning. With mature in-house technical tools at our disposal, we often deploy data driven DNN models to create the digital twins of the managed assets. The collection of DNN models deployed in Ceyark IIoT Platform will continuously injest field measurements over time from their respective managed assets and using commonly recognized methods to update the weights and biases of our DNN layers, the model gradually and over a period of time becomes capable and digitally identical to the managed asset. Computational process is then applied on the digital twin model to improve efficiency, identify mission critical issues, anomaly detection, reduce non-productive time, predict early signs of machinery fatigue well before they occur etc.
With Ceyark IIoT Platform, sensor measurements are propagated further from PLC to the inference server. The Ceyark IIoT Platform is a collection of different open source software components that are assembled together to provide an ability to ingest streamed data from different sensors through PLC systems, apply filter on incoming data, do data transformation like aggregation and finally push them to data store for long term storage. Ceyark IIoT Platform has all the necessary industrial integration blocks to integrate and collect streamed data from different sensor systems. Such streamed data are converted into metadata structures by the edge devices and sent to Analytics server. The platform Collect and analyze real-time device data to trigger automatic alerts and actions, including performing remote diagnostics and automatically initiating maintenance requests.

The Ceyark IIoT Platform is capable to handle massive amount of sensor measurements metadata and can be scaled horizontally to handle extra load on demand. The platform enables to connect and monitor all industrial assets using standard OPC-UA Server and extract data from factory floor devices to drive increased performance on the factory floor. Stored metadata will be used for analysis by applying predefined rules much in the same way an expert production manager would do. Analytics servers are more efficient in its ability to apply logic on vast amount of historical data to identify patterns of failure at near real time. Using stored metadata which were basically streamed  data from sensors and devices, we can use them to predict equipment failures and avoid costly repairs. This helps to go beyond monitoring factory assets and enables us to catch potential issues before they turn into problems.

The Ceyark IIoT Platform is made up of key modules.

• The IIoT Edge Device
  • With high bandwidth data processing and communication capability and ability to integrate with different vendor PLCs and support to large set of protocol standards

1. 1. 1. OPA UA
      2. Modbus devices (RTU, TCP, Ascii )
      3. BACnet
      4. M-Bus
      5. IO Link (low bandwidth devices)
      6. Ethernet TCP/IP devices (high bandwidth devices)
      7. Profinet
      8. DH+
      9. RS-485
      10. MQTT Devices

• • Service running Anamoly Detection Machine learning models to find unusual occurrences to identify and predict rare or unusual data points eg., Catch abnormal equipment readings.
• IIoT Edge Hub