Architecture To Bridge Physical world to Virtual Digital World
Introduction
How to realize an intelligent operation, smart manufacturing in an Industry and its physical assets is the challenge. The Industry were facing lot of challenges like, how to improve the efficiency, resiliency, throughput, quality etc. some of the global names in the manufacturing industry has been doing a research and proof of concepts on digitizing the process and machines behaviors on Industry 4.0.
Digital Twin platform is an effective means to reflect the physical status in virtual space. It breaks the barrier between physical world and digital world of manufacturing.
Digital Twins platform is one of the top 10 strategic technology trends named by Gartner Inc. 2017. Digital Twins represents the convergence of physical world such as Industrial product, household products will get dynamic digital representation.
The Digital Twin idea first evolved at NASA: full scale mockups of early space capsules, used on the group to mirror and diagnose problems in orbit, eventually give way to fully digital simulations.
Lockheed Martin, the world’s largest weapons manufacturer, listed Digital Twin as the top 6 of the top defense and aerospace industry.
China Association of Science and Technology Smart Manufacturing Academy Consortium listed Digital Twin as the one of the top 10 technological advancement in the world.
The world’s industry vision to be integrated with the Industry 4.0 principles has raised a lot of attention in the R & D community. It is focus on creating smart, networked world with smart and New IT technologies, procedures and process.
Industry 4.0 is only possible with Digital Twins — Dr. Michael Grieves
What are a Digital Twins?
“Digital Twin is a sensor enabled digital model of a physical object that simulates the object in a live setting.”- Dr. Michael Grieves.
A Digital Twins is a digital representation of physical world. The technical capabilities behind Digital Twins has expanded to include buildings, industries, people, processes, household etc.
It is essentially a computerized mirror of a physical asset and/or process, a virtual replica that relies on real-time data to mimic any changes that occur throughout the lifecycle.
The Digital Twins idea was first conceived by Michael Grieves at the University of Michigan in 2002. Due to technology unavailability, until the organization stores data on its own without using it, now is the right time to consider by evolution of technologies such as AI, ML, IOT, Cloud and Quantum computing.
Digital Twins is a vital software and tool to help engineers to understand not only how products are performing, but how they will perform in future. Analysis of the data from the connected sensors, combined with other source of information, allows us to make these predictions.
About the Article
This article provides details of what are they? Why are they needed? And how can they have developed?
You can find 100’s of Digital Twins articles in internet but few talks about the actual software, architecture used, design and skills that will happen. There are few article and reports explain clearly.
This article talks about what software’s to be used, the reference architecture and design and what skills require to develop a Digital Twin technology.
Evolution of Digital Twin
As stated, at present, the Digital Twin concept is more suitable to the Manufacturing Industry, smart home like machines, boilers, shopping floor etc, so, the details of next sections are more towards manufacturing Industry, so, shift your thinking towards shop floor etc.
The below diagram try illustrates and memorizes the evolution of Digital Twins in a Manufacturing Industry.
Initially the manufacturing materials are designed using 2D CAD drawings and part specifications, after few years, the 3D CAD models evolved, Bill of materials, bill process, industry process, industry representation and factory floors are become digital with software in place.
Later Industry houses extended the digitization across enterprises and the extended enterprises (partners, vendors etc.) and started moving towards the simulation of output, Hardware, software and manufacturing process.
In recent times, the evolution of sensors helps to connect all the manufacturing units with IOT, robotics and movement of real time data across units and units started acting based on the Hypothetical driven development.
Digital Twin concepts evolved with the advancement of IOT, Data Analytics and Microservices architecture and technologies.
Why Digital Twins required?
Digital Twin is a simulation model that represents a machine or a business process.
The Digital Twin will help manufacturing and business: Behavioral of machines with predictive and preventive analysis. Improve process and functioning, reduces the industrial accidents etc.
The below are the few major industrial accidents across globe.
By 2018, companies who invest in digital twin technology had seen a 30% improvement in cycle times of critical process
Industrial Accidents
The Industrial accidents are caused either by accident, negligence, incompetence or manual dependent. These accidents cause human and financial losses,
Highlights of few accidents (Wikipedia)
1. December 6, 1917: Halifax, Canada. The Halifax Explosion. A ship loaded with about 9,000 tons of high explosives destined of France caught fire as a result of a collision in Halifax harbor and exploded, killed around 2000 and 9000 injured
2. December 3, 1984: The Bhopal Disaster in India. A runway reaction in a tank containing poisonous methyl isocyanate caused the pressure relief system to vent large amounts to the atmosphere at a Union Carbide India Limited plant. Estimates of death 4,000 to 20,000.
3. October 23, 1989: Phillips Disaster, An explosion and fire killed 23 and injured 314
4. April 19, 2000: Pingxiang, Jiangxi, China, An Oxygen generator exploded at a steel factory killed at least 19.
5. August 23, 2016: Chittagong, Bangladesh, An incident of gas leakage happened at a fertilizer company in port city of Chittagong.
Etc. and many more… (Refer https://en.wikipedia.org/wiki/List_of_industrial_disasters)
Almost all accidents can be avoided if the plants or cities had Digital Twins Platform
How Digital Twins Helps?
· Visualizing products in use, by real users, in real-time
· Maintaining remote systems and machines
· Building digital connection, connecting disparate systems
· Provide accurate results with predictive, preventive and behavioral analytics
· Reduces asset downtime
· Increase Market Agility
· Reduce Cycle time
· Improve factory and plat efficiency
· Reduce maintenance and operational cost
Digital Twins give manufacturers and businesses an unprecedented view of their products performance. Digital Twins helps to identify potential faults, troubleshoot which avoid accidents and customer satisfaction.
The Digital Twins are software representation of assets and process that are used to understand, predict and optimize performance in order to achieve improved business outcomes.
General Electric Inc says, “By building a digital twin “model of one” for every critical turbine assembly, and continuously analyzing each model using advanced statistical tools, plant operators can bring turbines down for maintenance predictively, eliminating the costs of unnecessary downtime and mitigating the risks of unplanned outages.”
Data collected by IOT sensors onboard the physical assets. By applying advanced analytics and machine learning to rich, continuously improving asset models.
Uses of Digital Twin
· Asset behavior optimization by applying real time analysis to the virtual object and the modifying the behavior of the real object system
· Suggest the optimization to the real object system
· Observe the current real object system behavior and status by applying sensor readings to the Virtual object and observing its behavior
· Observe the historical behavior and the status of the asset
· Simulate the real object system which help to optimize the configuration
· Predict the future behavior by running predictive and behavioral analysis
Digital Twin Architecture
Digital Twin Communication
Each industrial machine consisted of two systems, the physical/real machines that always existed and a new virtual machine that contained all the information about the physical machine. This means that there was a mirroring of twinning of systems between what existed in real space to what existed in virtual space and vice versa.
The two Real and Virtual linked throughout the entire lifecycle of the system.
The flow goes as the Real object system built, up and running, the data from real space sent to virtual space. The virtual representation of that exact physical system is created in digital space.
Digital Twin Implementation
The Digital Twins can be implemented in multiple ways depending on the type of the Industrial machine you want to create a twin.
- Digital Twin Prototype (DTP): Digital Twin Prototype describes the physical artifact. It contains the informational sets and virtual version of real objects. This provides information such as 3D model, Bill of Material (BOM) with detailed specification (BOS), Bill of Processes (BOP), Bill of Services etc.
- Digital Twin Instance (DTI): Digital Twin Instance is a Digital Twin always linked to the Real system throughout the life of that physical product, it contains an exact 3D model, BOM, BOP, BOS etc. along with the results of any measurements and tests on the instances, a service record with past services and replaced components. Operational states capture from the actual sensor data in real machines.
- Digital Twin Aggregate (DTA): This is the aggregation of all the DTIs and captures the group of data structures from DTI. It queries all the data in DTIs and analyze together. It continually examines senor readings and correlate those sensor readings.
- Digital Twin Environment (DTE): This is the end to end environment setup to operate Digital Twins. The operations include receive data from real machines, analyze the data, predictive analysis, behavioral analysis etc.
Digital Twin Development
The development of Digital Twin is an iterative approach which involves multiple steps
First Step — The entire Real object process only on a physical version
Second Step — A digital version (virtual object) added with arguments of the physical version with additional information
Third Step — A real time interaction between Real object and Virtual object
Fourth Step — Detailed interaction between Real and Virtual object and virtual object start collecting real data and does a predictive, preventive, interrogative and behavioral analytics by applying Machine Learning and Deep Learning Algorithms.
Fifth Step — Start sending commands to the Real object (depends on the type of Industry Asset)
Digital Twin Components
Digital Twins are much more than graphical models. Machine Learning algorithms applied to production data detect correlation and prediction about life of Real object system.
Digital Twins consists of following main components
1. 3D Model — 3D Model connect, validate and optimize a physical object
2. IOT sensors — Sensors to be fit around the original system to get the data.
3. Data Model — Data Structure, metadata, managed instances of data elements
4. AI and ML enabled Analytics or algorithm — AI and ML model to analyze the sensor data and predict the future action.
5. Knowledge — Knowledge of unique physical thing.
High Level Architecture
A Digital Twin is
1. Represent an existing operational object — Design to decomissioning, the twin reflects the specific object
2. Represent the object’s real -world state — Provide data that describes present and historical condition.
Combined, these two create a virtual representation of a real-world system and its status. The update frequency of the data from the physical object will vary, the turbine, machine provide real time data asynchronously but supply chain provides data with latency.
Various kinds of analytics will be extracted from the data captured and provides the present state of the machine, the Twin may not calculate every state from day one, it can be iterative based on the state of the enterprise and notification will be generated based on the criticality.
The Physical Asset equipped with sensors that gather data which will be transferred over a network and actuators.
Gateways: Data move from physical asset to the Digital Twin and vice versa through gateways. A gateway provides connectivity between physical object and the Digital Twin, this enables the data pre-processing and filtering and transmits commands from Digital Twin to Physical Asset.
Data Stream: Effective transition of Physical asset data to a data lake and control application
Data Lake: It is used store the data generated by the physical asset.
Data Warehouse: Filtered and pre-processed data, it contains only cleaned data
Data Analytics: Data Analytics use data in Data Warehouse to generate Predictive, Behavioral and Interrogative analysis by using the Machine Learning Algorithms
Control Application: Send automatic commands and alerts to actuators
The Virtual object will be designed with Model Based Definition principle by using various tools.
Model Based Definition (MDB)
The MDB enables the production of a complete digital definition of a product a 3D model, thus replacing a traditional model.
The Digital Twin is a set of 3D model that connect, validate and optimize a physical/Real product, a process and a procedure. The objective is to design and run production systems that most efficiently process entities using the resources.
A model is defined by 1. Properties 2. States 3. Events 4. External View 5. Logic, these are the important factor to be considered when you design a Digital Twin.
Properties are input values that can be specified by the user of the physical object
States are dynamic values that may change as the model executes
Events are object may “fire” at selected items
The external view of the 3 D representation of the object
The Twin Model can develop by using various 3 D Model tools, ex: Simio, Seebo
These tools enable you to create a digital replica of your production asset, start with 2/3 D model and wire all the required twin component.
Simulate Digital Twin
Before industrializing the complete twin, create a simulation that interact with the digital twin model prototype you have created and validate your use cases, data flows etc. and simulate all events in your industry asset to trigger and test the events in your prototype. This help you to analyze how an object, product or system will work in the future after all the necessary changes are implemented.
The Simulation model contain complex algorithm to duplicates the operation of the asset and can diagnose unforeseen situations. The data is from sensors connected to the product or process are used to provide real-time boundary condition.
Gateway
The Gateway is to connect sensors to the WAN/Twin Environment, the sensors usually have very limited networking capability and utilized some protocols such as ZigBee or Bluetooth low energy. These protocols can’t directly connect to the twin environment, so Gateway helps you connect your Asset sensors to the Twin Environment.
The Gateway is not just Dumb Pipe to transfer data from sensors to Twin Environment, but it performs pre-processing of information in the Industry before sending data to the Twin Environment. You can monitor the sensors by using the Gateway.
Why Pre-processor data in Gateway?
The Analog data from sensors create large volumes of data, entire data may not be useful for further analysis in Digital Twin, may interested some few data like motor rotation, vibration, voltage etc., in some cases possible of 100’s of TB of data, this is high volume of data to send to Twin environment, so it is best to pre-process at the Gateway itself.
The Gateway is a hardware and software and consists of processor/microcontroller, sensors, protection circuity and connectivity modules.
The IOT Gateway enables
1. Connectivity to the Twin environment
2. Connectivity down to sensors in physical asset and existing controllers embedded in the system
3. Pre-process filtering of selected data for delivery
4. Local decision making, enabling easy connectivity to legacy systems
The IOT Gateway’s are likely build on open architecture to ensure interoperability between systems, enable wide application development.
Following are the various software stack used to develop Gateway
1. Connectivity (Protocols): MQTT, ZigBee, Bluetooth, USB, Wi-Fi Access Point
2. Runtime Environment: JRE, OSGI, .NET, Lua etc.
3. Security: Open SSL, Digital Signature, Certificate Management, Encrypted Storage, Monitoring, Audit Trail, Secret as a Service etc.
4. Management: Configuration as a Service (CaaS), etc.
All these software’s and protocols are run on the Linux platform.
IOT Edge
IOT Edge provide device connectivity and analytics to Physical asset in Factory IOT Hub environment within Digital Twin Environment. The data stored in the Factory IT Hub with the help of IOT Edge requires further immediate processing and analysis. The Edge IT hub sits in the same facility as Physical asset with sensors, because IOT data easily eat up network bandwidth and swamp your data center environment and resources. You use Machine Learning algorithm at the edge to scan for anomalies that identify impeding maintenance problems that require immediate action. With the ML, you could use visualization tools and techniques to show dashboards etc.
AI and Machine Learning in Digital Twin
The operation and optimization of complex factory systems requires modern, data and simulation driven multi-criteria decision approaches. The systems are very complex in terms of layout variability, control strategies, business process and system parameters. All these aspects are not independent, and these are co-existing. For these complex aspect’s decision, you require specific algorithm and data.
The monitoring and analysis of real time data allow improve digital twins design, control and strategies, by using the data, the digital twin can simulate virtual before applying to the real time object.
Data are the key for the success of Digital Twin and for any AI model use case. It is crucial to determine the type, quantity and quality of data and most important is the real-time data which is key.
· As mentioned above, the Digital Twins require real-time data from physical system in order to interface between the physical and digital world. you will get a lot of data from various system in enterprise, IOT sensors etc. but the key is quality and reliability of these data gathered from various source. You need to apply anomaly detection of data.
· Following are the few algorithms to be applied to predict the behavioral of the physical system/object.
o Physics based ML Algorithm: This algorithm used to analyses whether the data received from sensors truly from physical asset or out of calibration. If a sensor issue identified the physical based models can be used to provide a virtual sensor reading.
o Advanced Signal Processing: This technique is used to detect sensor noise and analyze and modify a signal. To analyze the signal, you need to use various algorithm such as Wavelet, kernel algorithm etc.
o Statistical process control: it is an industry methodology for measuring and controlling quality of sensor data of manufacturing process. The data is plotted on a graph with pre-determined control limits.
o Machine Learning and Deep Learning Algorithms: Various ML and DML Algorithms are used to detect the sensor data.
Physics based Machine Learning Algorithms
These models enable anomaly detection through the comparison of calculated parameters with measured values. The lack of system understanding from either experimental or computational results necessitated a different approach to predicting the real object behavior. The Physical ML Algorithm, which aimed to combine experimental and computational data.
Understanding the system behavior for knowing how it will react in various circumstances and modifying it for a specific purpose. For example, how the gas turbine manages the pressure and maintain and coordinating the pressure measurement across values is complex, these intuitive behavior with clearly correlated cause and affect relationship between various values. Predicting these complex systems can be very hard and inaccurate. Various technique can be used to predict the behavior of complex objects, but each has significant drawbacks.
One technique is to physically build and experimentally test a system to create data repository that can be referenced to predict the future behavior. By analyzing the expected values throughout the physical system with the reported values, the Neural Network model and Physical based ML to be used to understand the sensor data. They work to estimate the output of the system by using weights and biases by using Neural Networks.
How to utilize physics-based machine learning model uses physics, usually uses two approaches
1. Using physics theory, calculate additional features to feed into the model along with the measurements
2. Add physical inconsistency term to the loss function in order to punish physically inconsistent data
Advanced Signal Processing
To detect anomalies in the presence of sensor noise advanced signal processing technique can be used.
The Advanced Signal Processing addresses the development of new, advanced, signal processing method for studying the characteristics of highly non-stationary wave forms involved in the analysis, monitoring and management of sensor nodes.
This can be done by using ML Algorithms wavelets, Kernel regression and multi sensor data fusion technique.
Wavelet Algorithm
The Wavelet is a wave-like oscillation with an amplitude that begins at zero, increases and then decreases back to zero (Wikipedia), it is a mathematical function that decomposes a signal into a representation that shows signal details and trends as a function of time.
Wavelet transform is like Fourier transform with a completely different merit function. The Fourier transform is a powerful tool for data analysis but does not represent abrupt changes efficiently because it is not localized time and space, so you need to localized time and frequency, so wavelet does this.
The wavelet exist for a final duration and waves are come in different sizes and shapes. Following are the two key concepts of Wavelet used to analyze the data
1. Scaling and 2. Shifting
The scaling is stretching and shrinking of signal in time from sensors and can be expressed by using this equation , s is the scaling factor and corresponds to how much signal scale in terms of time and scale factor inversely proportional to frequency. the bigger scale factor in the wave result in low frequency and small-scale factor of wave signal result higher frequency from sensors.
The stretched wavelets help to capture the slowly varying changes in sensor signals and compressed wavelet helps in capturing the abrupt changes of the signals from sensors.
Shifting is the delaying the wavelets along with the length of the signals.
Kernel Regression
This technique is to estimate the conditional expectation of a random variable. The objective is to find a non-linear relation between a pair of random senor variables x and y.
Statistical Process Control
The control charts are used to monitor the quality of the sensor data, depending on the process characteristics to be monitored, you can use various charts such as Univariate and multivariate control charts.
The control chart shows the value of the quality characteristics versus the sample number or versus time. The chart contains the central line that represent the mean value. All the data points from IOT sensors are fall between Lower control limit and upper control limit.
Univariate control chart: This can be used to display chart for one quality characteristics
Multivariate Control Chart: This can be used to display chart for more than one quality characteristics.
Machine Learning and Deep Learning Algorithms
Algorithms such as logistics regression, decision trees, random forest methods, clustering methodologies and neural networks are used in Digital twins.
Digital Twin Architecture Principles
Digital Twin Identification: Types representing a Digital Twin must be identifying properties such as name and twin type. Each twin must be unique name and type is a class of digital twin (ex: engine, gas turbine etc)
Loosely coupled relationships — Represent data is only loosely connected
Local data analytics: setup the digital twin architecture local to the factory for immediate analysis and move to cloud for broader analysis
Data Collection: Data are read from various sensors and collected in twin set and architecture require to transform data to usable form
Enrich data: Collected data require to enrich based on the interconnected data from other sensors.
Modeling: Prepare a 100% accurate virtual model to simulate the physical model
Enterprise Integration: Analyze the integration point in an enterprise as digital twin need to get interconnected data from various systems in an enterprise
Conclusion
Increasing digitalization on every stage of manufacturing is opening opportunities for manufacturers to achieve a whole new level productivity. Autonomy provides the production system with the ability to respond to unexpected events in an intelligent and efficient manner without the need for re-configuration at the supervisory level. Lastly ubiquitous connectivity such as the IOT facilitates closing of digitalization loop, allowing next cycle of product design and production execution to be optimized for higher performance.
Design a Digital Twin with care and simulate exact model and setup the Digital Twin in Factory space for immediate analysis and send data to cloud for further analysis.
Collect the quality data as IOT sensors generates TB of data, apply data cleansing method to clean and apply ML algorithm for further analysis.
Real time and notification is very important for avoiding unnecessary incidents in Physical object.
References
Statistical Controls: https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc31.htm
Leveraging Digital Twin Technology in Model-based System Engineering — https://www.mdpi.com/2079-8954/7/1/7/htm
Industry 4.0 and Digital Twin Model — https://www.simio.com/index.php
Digital Twins for Decision Making in complex production and logistics environment
Development of Physics Based Machine Learning Algorithms
Digital Twin — The new face of innovation
RedelinghuysBassonKruger_SixLayerDigitalTwinArchitecture20180506a