Tools that enhance human capabilities and flexible manufacturing systems are necessary to deliver on the core promises of the 4th industrial revolution.

Industry 4.0 Deep Dive. Part 5: Flexible manufacturing

<p>For many years, silicon Physical Unclonable Functions (PUFs) have been seen as a promising and innovative security technology that was making steady progress. Today, Static Random-Access Memory (SRAM-)based PUFs offer a mature and viable security component that is achieving widespread adoption in commercial products. They are found in devices ranging from tiny sensors and microcontrollers to high-performance Field-Programmable Gate Arrays (FPGAs) and secure elements where they protect financial transactions, user privacy, and military secrets.</p><h2>Physical Unclonable Function (PUF)</h2><ul><li>A PUF provides a way to authenticate electronic devices or chips and generate cryptographic keys from their unclonable silicon fingerprint.</li><li>It uses uncontrollable deep-submicron variations from manufacturing to derive a stable, unclonable, and unique device identity.</li></ul><h2>Benefits of SRAM PUF</h2><ul><li>No need to program keys or obtain entropy externally.</li><li>No secrets stored in device when power is off.</li><li>Keys and identifiers based on unclonable silicon structure.</li><li>Uses existing, standard hardware (SRAM) as PUF.</li></ul><h2>Reliability and Certification</h2><ul><li>Advanced error correction.</li><li>Based on years of extensive reliability testing across technology nodes and operating environments.</li><li>Devices using SRAM PUFs have been government vetted and certified EMVCo-Visa and CC EAL6+.</li></ul><h2>Application Markets</h2><ul><li>Smartcard, FPGA</li><li>Sensors, microcontrollers</li><li>Mobile, automotive, IoT</li></ul><h1>The SRAM-based PUF</h1><p>Due to deep submicron manufacturing process variations, every transistor in an Integrated Circuit (IC) has slightly different physical properties. These lead to small but measurable differences in terms of electronic properties, such as transistor threshold voltages and gain factor. Since these process variations are not fully controllable during manufacturing, these physical device properties cannot be copied or cloned.</p><p>Threshold voltages are susceptible to environmental conditions such as temperature, and voltage so their values cannot be used directly as unique secret keys or identifiers.&nbsp;</p><p>The behavior of an SRAM cell, on the other hand, depends on the difference of the threshold voltages of its transistors. Even the smallest differences will be amplified and push the SRAM cell into one of two stable states. Its PUF behavior is therefore much more stable than the underlying threshold voltages, making it the most straightforward and most stable way to use the threshold voltages to build an identifier.</p><h2><span class="fr-img-caption fr-fic fr-dib" style="width: 643px;"><span class="fr-img-wrap"><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1591264573768-SRAM_behavior2.png"><span class="fr-inner">Figure 1: Extracting a strong secret key from SRAM behavior.&nbsp;</span></span></span></h2><h2>SRAM PUF Behavior</h2><p>An SRAM memory consists of a number of SRAM cells. Each SRAM cell consists of two cross-coupled inverters that each are built up by a p- and n-MOS transistor. When power is applied to an SRAM cell, its logical state is determined by the relation between the threshold voltages of the p-MOS transistors in the invertors. The transistor that starts conducting first determines the outcome, a logical &#39;0&#39; or &#39;1&#39;.</p><p>It turns out that every SRAM cell has its own preferred state every time the SRAM is powered resulting from the random differences in the threshold voltages. This preference is independent from the preference of the neighboring cells and independent of the location of the cell on the chip or on the wafer.</p><p>Hence an SRAM region yields a unique and random pattern of 0&#39;s and 1&#39;s. This pattern can be called an SRAM fingerprint since it is unique per SRAM and hence per chip. It can be used as a PUF.</p><p>Keys that are derived from the SRAM PUF are not stored &#39;on the chip&#39; but they are extracted &#39;from the chip&#39;, only when they are needed. In that way they are only present in the chip during a very short time window. When the SRAM is not powered there is no key present on the chip making the solution very secure.</p><h2>PUF Reliability</h2><p>The deep submicron process variations that determine PUF behavior are frozen during manufacturing and do not change afterwards. Hence the preference of the SRAM cells is persistent and stable over time.</p><p>However, there is still a degree of noise. A small number of the cells, close to equilibrium are unstable and display a seemingly random startup preference. So, each time the SRAM starts up, a slightly different pattern emerges. This noise component is dependent on temperature, voltage ramp, and operating conditions.</p><p>The noise of SRAM-based PUF responses has been exhaustively characterized and tested under a wide variety of circumstances and foundry processes:</p><ul><li>Temperatures ranging from -55&deg;C to +150&deg;C (-67&deg;F to 300&deg;F)</li><li>Voltage variation +/-20%</li><li>Humidity up to 80%</li><li>EMC tests at 3V/m (EN55020 0.15&ndash;150 MHz and IEC 61000-4-3 80-1000MHz)</li></ul><p>In particular, the SRAM PUF has been qualified for automotive, industrial and military use in collaboration with customers and partners.</p><p>Millions of measurements have been performed. Under all these circumstances the average noise level of the SRAM-based PUF response was found to be lower than about 15%. Despite this amount of noise, it is possible to reconstruct a high-entropy device unique and reliable key every time the SRAM is powered. This can be done by applying error correction techniques, such as &#39;helper data algorithms&#39;<sup>[1]</sup> or &#39;fuzzy extractors&#39;<sup>[2]</sup>. These algorithms perform two main functions that will be explained below: error correction and privacy amplification.</p><h2>Error Correction</h2><p>Error correction techniques for cryptographic key reconstruction require an enrollment phase and a reconstruction phase. In the enrollment phase (a one-time process) the PUF response is mapped onto a codeword of an error correcting code. Information about the mapping is stored in the Activation Code (AC) or helper data. The AC is constructed such that it does not leak any information about the key. It should be stored in memory that is accessible by the PUF algorithms, but it can be stored off-chip as it is not sensitive. Any change to the AC, malicious or not, will prevent key reconstruction. The AC is only valid for the chip on which it was created.</p><p>Each time the device runs an authentication protocol and needs the secret PUF key, a new noisy PUF measurement is carried out and the PUF key (without noise) is extracted from the AC and this new PUF response. This is called the reconstruction phase. Both enrollment and reconstruction phases are illustrated in Figure 2.</p><p>The error correction algorithms have been designed to reconstruct the key with an average error rate of less than 10<sup>-12</sup> and an error rate of less than 10<sup>-9</sup> under extreme circumstances of 25% average noise.</p><h2><span class="fr-img-caption fr-fic fr-dib" style="width: 707px;"><span class="fr-img-wrap"><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1591264692705-Enroll_reconstruct.png"><span class="fr-inner"><span contenteditable="true">Figure 2: Enrollment and reconstruction phase for the generation of PUF keys. Note that R is the initial PUF response during enrollment while R&#39; is a PUF response in the field with a noise component.</span></span></span></span></h2><h2>Privacy Amplification and Security</h2><p>Secret keys provide security based on the fact that they are completely random and hence unpredictable. Physical measurements, such as PUF responses, have a high degree of randomness, but are usually not completely uniformly random. Privacy amplification is used to generate uniformly random keys.</p><p>By combining error correction and privacy amplification, a 1kByte SRAM PUF response can be turned into a 256-bit uniformly random key and only approximately 0.5 kByte is needed for a 128-bit key with full randomness. A typical SRAM PUF contains so much entropy that only a few dozen bytes are needed to provide a collision-free globally unique identifier that can be used as a unique (but noisy) electronic chip ID (ECID) or as a serial number.</p><p>Dedicated security labs and security teams at customers have analyzed the security of Intrinsic ID&#39;s SRAM PUF against various invasive and non-invasive physical attacks without revealing any weaknesses. Attacks with Scanning Electron Microscopes, Lasers, FIBs and probes have not been successful. Side-channel attacks have not led to any leakage of sensitive information.</p><h2>Aging</h2><p>Accelerated aging tests have been performed on SRAM PUFs to investigate the noise level as a function of time. By using a patented anti-aging technique<sup>[3]</sup> a 25-year lifetime can be guaranteed for SRAM PUF technology.</p><h1>SRAM PUF Implementation &ndash; QuiddiKey &amp; BroadKey</h1><p>Intrinsic ID has bundled the above error correction, randomness extraction, security countermeasures and anti-aging techniques into its products. They extract cryptographic keys from the SRAM PUF in a very secure manner and are available as Hardware IP (netlist) called QuiddiKey&reg;, firmware (ANSI C Code) called BroadKey&trade;, or a combination of these by using BroadKey&#39;s hardware interface. These hybrid solutions combining hardware and software offer great efficiency when already integrated hardware accelerators, such as for (a-)symmetric cryptography, are used in combination with SRAM PUF technology.</p><p>The hardware IP is small and fast &ndash; around 25K gates / 50K cycles &ndash; and connects to common interconnects, such as AMBA&reg; AHB, APB as well as proprietary interfaces. A Built-In Self-Test (BIST), diagnostics, and health checks are included in the logic. A driver is provided to ease integration with software. Since it is pure digital, single clock logic it synthesizes readily to any technology.</p><p>Software reference implementations start from 4KB of code and are available for major platforms, such as ARM&reg;, ARC&reg;, Intel&reg;, MIPS and RISC-V. Software implementations can be used to add PUF technology to existing products by a firmware upgrade. Intrinsic ID also provides versions of BroadKey that are pre-integrated with ARM&reg; TrustZone&reg;.</p><p>Both QuiddiKey hardware and BroadKey software solutions can be optimized for low footprint, low latency or low memory use depending on the application. Re-using or integrating with existing crypto cores and random number generators can further enhance performance and reduce footprint. Intrinsic ID solutions come with comprehensive product specifications and integration guidelines including reference code that illustrates usage of the API offered to the application programmer.</p><h2>Requirements</h2><p>These Intrinsic ID products use uninitialized SRAM. This can be a separate SRAM block or a part of a bigger existing SRAM. Standard SRAM suffices. To store the Activation Code (AC), access to a storage medium is needed. This can be embedded Non-Volatile-Memory (NVM), a separate memory on the board, e.g. flash, or cloud storage. For the firmware version BroadKey, a microcontroller is needed for which a C-compiler exists. The PUF algorithms can be stored in any NVM e.g., flash, ROM.</p><p>Note that SRAM is embedded in almost any microprocessor and SoC, in every technology node and is part of the standard manufacturing process. There is no need for time-consuming qualification and chip testing, as extensive testing by Intrinsic ID and its partners has shown that technology reliably scales down to the smallest technology nodes currently available.</p><div id="ftn1"><h2>Use in the Field</h2><p>SRAM-based PUFs are commercially available from semiconductor companies. They are already deployed in a wide range of microcontrollers, FPGAs and smart card controllers. In other markets, software implementations<sup>[4]</sup> have enabled rapid deployment of this technology even as a retrofit solution. Intrinsic ID has partnered with leading semiconductor companies and developed solutions for protecting embedded systems, sensors, and controllers. More information about our SRAM PUF deployments can be found on our website: <a href="https://www.intrinsic-id.com" target="_blank">https://www.intrinsic-id.com</a>.</p><h1>Conclusion</h1><p>SRAM-based Physical Unclonable Functions have been successfully implemented in commercial products. SRAM PUFs combine high security and reliability with low cost, low footprint designs, and easy implementation. They have been deployed in many devices, from tiny microcontrollers and sensors to high performance FPGAs and secure elements.</p><p>Numerous implementations have consistently demonstrated the reliability and security of the technology. SRAM PUFs are a mature and robust technology, designed for security and based on solid theoretical foundations. SRAM PUFs have gained their credibility in high security markets and are now gaining traction in markets ranging from low-cost IoT applications to high-end security solutions for government, defense, and the payment industry.</p><div id="ftn1"><h2>References</h2></div></div><p>[1] J.-P. Linnartz and P. Tuyls, &quot;New shielding functions to enhance privacy and prevent misuse of biometric templates,&quot; in International Conference on Audio and Video-based Biometric Person Authentication (AVBPA&#39;03), ser. LNCS, J. Kittler and M. S. Nixon, Eds., vol. 2688. Heidelberg: Springer-Verlag, 2003, pp. 393&ndash;402.</p><div id="ftn2"><p>[2] X. Boyen, &quot;Reusable cryptographic fuzzy extractors,&quot; in ACM Conference on Computer and Communications Security (CCS&#39;04). New York, NY, USA: ACM, 2004, pp. 82&ndash;91. AND Y. Dodis, L. Reyzin, and A. Smith, &quot;Fuzzy extractors: How to generate strong keys from biometrics and other noisy data,&quot; in EUROCRYPT&#39;04, ser. LNCS, C. Cachin and J. Camenisch, Eds., vol. 3027. Heidelberg: Springer-Verlag, 2004, pp. 523&ndash;540.</p><p>[3] R. Maes and V. van der Leest, &quot;Countering the effects of silicon aging on SRAM PUFs&quot;, Proc. IEEE Int. Symp. Hardw.-Oriented Secur. Trust (HOST), pp. 148-153 available at <a href="https://www.intrinsic-id.com/wp-content/uploads/2017/05/PUF_aging.pdf" rel="noopener noreferrer" target="_blank">https://www.intrinsic-id.com/wp-content/uploads/2017/05/PUF_aging.pdf</a>&nbsp;</p><p>[4] H.Hodson &ndash; New Scientist, &quot;Silicon fingerprint on chips could make any gadget unhackable&quot; June 6, 2016 <a href="https://www.newscientist.com/article/mg23030771-400-physical-quirks-in-silicon-chips-are-key-to-unhackable-devices/" rel="noopener noreferrer" target="_blank">https://www.newscientist.com/article/mg23030771-400-physical-quirks-in-silicon-chips-are-key-to-unhackable-devices/</a> or &quot;Chip design quirks make our lives more secure&quot; in the June 11 printed issue page 24.</p></div>

Low-cost and strong secret key storage technology is critical today to enable affordable and effective security systems.

SRAM PUF: The Secure Silicon Fingerprint

The fourth industrial revolution and the technologies that create it explained: The whole is bigger than its parts.

Industry 4.0 Deep Dive. Part 1: Introducing smart manufacturing

<p>Up to 30% of a compressor&#39;s normal output can be lost through leakage, and poorly maintained equipment will generally be leaking compressed air at around 20% of its total capacity. Leaks may also cause your whole system to lose pressure, leading to less efficient functioning of pneumatic devices, and loss of production output.</p><p>Not just the compressor, but almost all your system equipment will have its lifespan reduced, as leakages force more frequent cycling and consequent wear. To estimate how much your pneumatic system may be leaking will require running the compressor for a specified time when it is not driving any devices, and measuring the average time it takes to load and unload the system. This loading and unloading occurs as leaks cause pressure drops in the system, and the compressor then cycles on and off to compensate.</p><p>Not all leakages can easily be prevented, and system design should incorporate some standard allowance for a pressure drop rate per unit time. These parameters could be anticipated in the circuit design, so that appropriate pressure rates and reservoir sizes are determined.</p><h2>How To Find Leaks In Pneumatic Systems</h2><p>Sometimes, unnecessary additional compressor capacity is installed to compensate for pressure lost due to leakage. A better and more cost-effective alternative is to introduce a proactive programme of detection and repair, which can bring down your average leak rate to under 10%. The most common areas where leaks can occur include joints, couplings, seals and fittings, as well as hoses and areas which have missing welds or worn packing. Missing O-rings are a common problem, as are improperly installed FRL units and worn point-of-use tools.</p><p>Air is invisible, of course, so once you&#39;ve calculated the amount of air you&#39;re losing through leakage, you&#39;ll have to find some way of detecting the specific locus of the leak(s). A modern solution to this is ultrasonic detection, by means of an acoustic device to identify the high frequency sounds made by leaking air.</p><p>Since this type of equipment is commonly portable, it makes the detection of leaks very fast, accurate and versatile. Ultrasound is directional in its transmission, emitting a louder signal the closer it is to the source, so leak sites can be very easily pinpointed by scanning the general area where the equipment is located. Tests can be carried out while equipment is in use, and the detectors are capable of finding many different types of leak. This kind of device is very simple to operate, and it takes only a few minutes to learn how to use it.</p><h2>What Causes Leaks In Pneumatic Systems?</h2><p>Leaks are most often found at junction points between the system and end-use pneumatic applications, especially around connectors and joints. Fixing a leak can mean a straightforward tightening-up of a connection, or more complex replacements of faulty components like hoses, fittings, couplings, joints, pipe sections, traps or drains.</p><h2>Faulty Connectors</h2><p>Connectors are a very large family of components used to join hoses, tubes or peripheral devices. They need to be chosen carefully and precisely to ensure that a leak-proof joint is achieved, and sealed where necessary. Push-in connectors are very popular due to the simplicity of their installation, but they may be prone to suffer more leakage if a cycle involves frequent connection and disconnection.</p><h2>Pressure</h2><p>Leakage may also ensue if the pressure load on valves or connectors exceeds their specified rating. The flow rate of air depends on the pressure differential pertaining across a leakage point, so a lower overall system pressure will help reduce the rate of leakage. If you can stabilise the header pressure of the system at the minimum range that is practical, the leakage rate should also be minimised.</p><h2>Too Many Connection Nodes</h2><p>However carefully you seal your connections, if they are pushed to function outside their pressure parameters, some leakage is bound to occur. Furthermore, the more connections there are in the system, the greater the likelihood of multiple leakages leading to a cumulative drop in overall pressure. Increasing the number of connections is not therefore a cost-effective way to expand your pneumatic system, as it also makes the system more complicated, and more difficult to detect leaks.</p><h2>Wrong Connection Type</h2><p>Thread types come in many varying styles and sizes, and tend to differ according to their country of origin and use. Connections typically use male and female mating parts, so it&#39;s important to make sure that the thread types match. There are also recommended torque values for most threaded joints, and connections should not be over or under-tightened in relation to these values. Threaded joints should be secured with teflon tape.Seal Degradation</p><p>Exposure of some materials to chemicals may make them degrade. Items such as rubber washers may develop cracks that allow leakage, or thread sealant might be of the wrong type, or not properly applied. All fittings, tubing, hoses, disconnects, etc, should be of high quality, and properly installed with the right thread sealant.</p><h2>Cylinder Bore Wear</h2><p>It&#39;s possible for a pneumatic cylinder to suffer excessive wear on the bore if it doesn&#39;t have the proper lubricant. Misalignment or improper mounting can also cause lateral forces to be applied on the bore, which could cause air to leak along the piston&#39;s periphery.</p><h2>Predictive Leak Prevention</h2><p>Establishing a good leakage identification and repair programme is critical to maintaining a stable, reliable, efficient and cost effective pneumatic system. Such a programme should involve employee teams in the identification and tagging of leaks, tracking performance, carrying out repairs and verification. Whenever leaks are identified and repairs instituted, the whole system should be reassessed. Scheduled predictive maintenance can help you detect and deal with leaks, so that you save on reactive maintenance and unscheduled downtime. An aggressive programme of leakage prevention and reduction should therefore form part of the overall performance improvement of pneumatic systems.</p><hr><p>This post was originally posted on&nbsp;<a data-fr-linked="true" href="https://www.rowse.co.uk/blog/post/how-to-find-leaks-in-pneumatic-systems" rel="noopener noreferrer" target="_blank">www.rowse.co.uk</a>.</p>

If a compressed air system develops a leak, it can waste a lot of energy, especially in a complex automated industrial setting.

How To Find Leaks In Pneumatic Systems

<p>Automated systems governing smart manufacturing, and enabled by IoT devices, give us the IIoT (Industrial Internet of Things), which allows for the expansion and streamlining of all aspects of business. In addition, the deployment of AI and robots is particularly useful in industry as they increase product quality, production efficiency and overall speed.</p><p>The introduction of collaborative robots (or cobots) is becoming increasingly common, in industry as well as in laboratory work and commerce. Cobots can take instructions from humans in order to work more productively with them, including instructions that were not originally anticipated in the initial programming. Robots don&#39;t get bored, hungry or tired. The <a href="https://www.rowse.co.uk/blog/post/the-rise-of-artificial-intelligence" target="_blank" title="The Rise of Artificial Intelligence">capabilities of Artificial Intelligence</a> are far beyond human capacity when it comes to such things as miniaturisation and precision measurements, and they deliver vastly superior quality assurance.</p><p><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1584368176061-1584368176061.png" style="width: 1080px;" class="fr-fic fr-dib"></p><p>So what are the benefits we can expect to see as industry moves forward into the 21st century world of cobots and AI?</p><h2>1. Direct Automation</h2><p>IIoT connects all IoT enabled devices to the factory floor, integrating manufacturing processes with big data and making them programmable via a logic controller. Increased use of precision sensory equipment means that information can be generated, recorded and analysed for all aspects of the production process, covering anything from temperature to item picking and packaging. Programmable logic controllers with AI capacity for deep learning can then respond automatically to the seamlessly generated information, and make alterations to the minutest function without recourse to human intervention. The big data analytics processed by AI can substantially improve performance across the entire production process, and can be operated remotely.</p><h2>2. 24/7 Production</h2><p>Human beings are biological organisms and need regular maintenance, e.g. food and sleep. For any production facility to continue working round the clock it is necessary to introduce shifts, using three human workers for every 24 hour period. Robots don&#39;t get tired or hungry, and are capable of working on the production line 24/7. This allows the expansion of production capabilities, which is increasingly necessary to meet the demands of worldwide customers. Furthermore, robots are more efficient in many areas such as the assembly line, and picking and packing departments. They can greatly reduce turn-round times in many areas of the business operation.</p><h2>3. Safety</h2><p>Human beings are fallible and prone to making mistakes, especially if they are tired or distracted. Errors and accidents do occur on the factory floor and in any construction or processing environment; a tendency which can be all but eradicated by AI and robotic assistance. Remote access control means a reduction in human resources, especially when the work is dangerous or requires superhuman effort. Even regular working environments will cut down on the incidence of industrial accidents and lead to an overall improvement in safety. In addition, more advanced sensory equipment integrated with IIoT devices make the installation of safety guards and barriers a simpler and more effective measure to protect human lives.</p><h2>4. Lower Operational Costs</h2><p>Many companies are viewing the introduction of AI into the manufacturing industry with trepidation, as it requires a huge capital investment. On the other hand, the ROI is significant and increases as time goes on. Once intelligent machines begin to take over the daily activities of a factory floor, businesses will benefit through considerably reduced operating costs, with predictive maintenance helping additionally to reduce machine downtime.</p><p>These days, consumers are increasing their demand for unique, personalised or customised products, while continuing to expect the best value. With such advances as 3D printing and IIoT connected devices, it is becoming simpler and cheaper to meet these needs, and using virtual or augmented reality design techniques means that the whole production process will be more cost-effective. Integrating machine learning and CAD means that systems can be designed and tested in a virtual model before they are put into production, thus reducing the cost of trial-and-error machine testing.</p><h2>5. Greater Efficiency</h2><p>IIoT enables the collection of vast amounts of data and advanced analytics which can be used to gain insights into consumer behaviour. Trends can be predicted, patterns recognised and market developments forecast across time, socioeconomic sectors and geographic markets, as well as taking into account political developments, macroeconomic cycles and even weather patterns. AI has the capacity with machine learning to anticipate information, to refine processes, and to track incongruities, all the way down the supply chain from source to finished product. This is particularly aided by such technologies as&nbsp;<a href="https://www.rowse.co.uk/blog/post/rfid-in-manufacturing" target="_blank" title="RFID In Manufacturing">RFID tracking</a>, which enable materials to be tracked without having to go through a physical process such as a bar code reader.</p><h2>6. Quality Control</h2><p>AI is also extremely useful for carrying out predictive maintenance on machinery and equipment. Using sensors to track performance and operating conditions, machines can learn to predict malfunctions and failures, and take action to remedy them before they occur. This can result in faster feedback, helping companies to eradicate unplanned downtimes.</p><p>Sensors can also detect the mot microscopic defects, scanning them at resolutions far beyond the capacity of human vision, thus improving productivity and increasing the percentage of items that will pass quality control. AI helps to speed up many routine processes and improves accuracy to an enormous extent. This obviates the requirement for quality control and in-process inspection by human beings, which is time-consuming and often fallible.</p><h2>7. Quick Decision Making</h2><p>When IIoT is coupled with cloud computing and virtual or augmented reality, companies can share simulations, confer on production activities and exchange critical or important information in real-time, irrespective of geographical location. The data gathered from sensors and beacons helps to determine consumer activity, allowing companies to anticipate future needs and make rapid decisions on production, as well as speeding up the exchange between manufacturers and suppliers.</p><hr><p>This post was originally posted on <a data-fr-linked="true" href="https://www.rowse.co.uk/blog/post/7-manufacturing-ai-benefits" rel="noopener noreferrer" target="_blank">www.rowse.co.uk</a>.</p>

As the world becomes increasingly automated, the IoT (Internet of Things) is already transforming our domestic and business lives. Nowhere is this more apparent than in the use of AI and robotics in the manufacturing industry, with all the benefits offered by Industry 4.0.

7 Benefits of AI In Manufacturing

<p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="f2d4">Most robots used in factories today are powerful but blind. They are programmed to do specific tasks repeatedly with high precision and speed, but until now have not been able to adapt to changes in the environment or handling variables. Robot arms often lack the ability to grasp an object if it is moved a few inches, not to mention picking or sorting millions of different products in a warehouse.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="1c10">As a result system integrators design shaker tables, feeders, conveyor belts, etc., so robots can be offered the same parts in the right orientation. Larger industrial robots need cages so they don’t accidentally hurt nearby people. Unlike human hands, which can perform tens of thousands of different tasks, robot arms can only be programmed for single tasks, and they need the proper end effectors.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="43d5">Additionally, most robots cannot sense, unless you equip them with force or vision sensors. They don’t know if tasks are performed properly, because they cannot see or feel.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="75cb">Traditional computer vision requires pre-registration of all objects, i.e. scanning and building computer-aided design (CAD) models for all items beforehand with expensive 3D cameras. The process is time-consuming and inflexible. Even if there was a box with exactly the same shape, but a different size, a robot couldn’t automatically recognize the differences and figure out how to grasp the new item.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="85ad">In addition, translucent packaging, reflective surfaces, and deformable objects (such as food and clothing) make robotic vision an even more difficult problem to solve. The additional non-robot hardware and integration end up costing four times more than a robot arm, which costs anywhere between $10,000 to $100,000.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="9cf8"><strong class="lj lv">Adding intelligence to robots</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="41fb">To solve these issues, deep reinforcement learning (DRL) is unleashing the full potential of robotics, giving them the ability to recognize and react to their surroundings and handle product variations. Robots can now learn from experiences with off-the-shelf cameras and deep neural networks.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="7081">With enough data and practice, robots can now teach themselves new abilities, as was seen with robots’ ability to identify images and win video games. With every grasp and trial, a robot gets smarter and more adept at mastering the task. Furthermore, cloud-connected robots allow learnings to be shared with each other. This is a tremendous shift, making robotics solutions more dexterous, flexible, and scalable.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="0fae"><strong class="lj lv">Deep Learning</strong> (DL) is primarily used for image classification. Deep neural networks learn representations, rather than task-specific algorithms.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="58a9"><strong class="lj lv">Reinforcement learning</strong> (RL) is inspired by behavioral psychology. It learns control policies by trial-and-error through reward or punishment.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="ba32"><strong class="lj lv">Deep Reinforcement Learning</strong> (DRL) combines these two approaches. It is currently suited for:</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="d407">· Simple or simulated tasks</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="0cad">· Fault-tolerant tasks</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="bf6e">· Just enough variation to make explicit modeling difficult</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="b81f">· Easy-to-define goals (reward functions)</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="25e2">· Semi-structured environments</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="ca31"><strong class="lj lv">Warehouse as a testing ground for AI</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="2bb1">Automation in warehouses is widely considered the “low-hanging fruit” for these technologies because tasks tend to be similar across warehouses, and the environment is more structured. For example, order picking represents more than 40% of operational costs across most warehouses. While tasks are repetitive, they involve some variation with products and packaging design, making it impossible to hard-program robots in advance.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="bf53">Furthermore, labor costs account for up to 70% of a warehouse’s total budget. Pressured by e-commerce companies such as Amazon, retailers are looking to automation to address labor shortages and reduce costs. Customer demand, price competition, and fast delivery requirements are driving a strong business case for warehouses to be the first adopter of AI-enabled robots.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="d7b2"><strong class="lj lv">Chart: Why do we see AI-enabled robots in the warehouse first?</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph=""><strong class="lj lv"><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNTY4OTAzMzQ5MzI3LTFfTDhkY3A1c1lfcEU0MTdqdW9VaEMwZy5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 766px;" class="fr-fic fr-dib"></strong><br></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="76b9">Most tasks in a warehouse are fault tolerant, unlike tasks on a production floor or in the automotive industry. If a robot accidentally drops an object in a warehouse, it can try again as long as it achieves the required speed and success rates without damaging the goods.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="d8c5">DRL is particularly suited for fault-tolerant simple tasks — actions a human can perform within a few seconds — with enough variation in semi-structured environments. As technology improves in environments like warehouses, it can then be applied in other industries, such as manufacturing, where tasks are more complicated and more uncertainties exist. Thus, an automated warehouse acts as an incubator, predicting a knowledge spill-over to many other market segments.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="36d8"><strong class="lj lv">Piece-picking is the holy grail</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="2047">Warehouse automation has been around for decades, in the form of automated storage, retrieval systems, and mobile robots like those from Kiva (now Amazon Robotics), that move items around.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="f719">But most of the pick-and-pack operations within a warehouse are still performed by humans, with labor costs accounting for 50 to 70% of the overall warehouse budget. Worker shortages have driven up salaries by a rate of 6 to 8% every year, while robot costs have fallen considerably since the 1990s.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="62e4">Piece-picking has long been the holy grail of robotics. Year after year, incumbents such as Amazon and KUKA host robotics challenges for startups and academic teams to create machines that can identify, pick up and stow goods. The emergence of deep learning allows robots to recognize, pick, and place hundreds or up to thousands of items.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="e9b0">But the technology is not yet perfect. It’s still challenging for a machine to recognize tens of millions of objects, and manipulate deformable objects or items with transparent packages. However, our interviews with Locus Robotics and Osaro indicate that many industry experts expect the technology will mature. In addition, we will have a robot vision camera that can recognize almost everything in a warehouse within five years. If that happens, it will affect not just warehouses, but retail, delivery, and many other applications that we haven’t yet imagined.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="9aae"><strong class="lj lv">Pushing towards standard models</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="68ce">Technology improvements in warehouse automation and beyond will also push the mature, but fragmented, robotics industry to become more standardized and modular. Given the nature of narrow AI, most Silicon Valley startups in the AI-enabled robotics industry choose to focus on specific vertical applications and collect proprietary data as a defensive advantage.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="155b">Piece-picking by robots is the best example. As the technology improves, we will see warehouse piece-picking startups enter manufacturing, food, and agriculture markets. Moving away from manually programming a robot and having it learn and act autonomously on specific simple tasks — and sharing the learning — will become more important in the future.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="a46c">However, it’s not easy to find scalable and technically feasible business cases like piece picking that customers are willing to pay for. For example, assembly tasks vary from factory to factory, making it difficult to get sufficient training data or scale a solution across different customers.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="b207">Given current technology constraints, there’s still a trade-off between precision and flexibility. Deep learning allows robots to handle variation, but these AI-enabled robots still cannot achieve the same precision and accuracy as traditional robots, at least not today.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="d366">Some startups are providing modules or technology stacks for other players in the ecosystem to mitigate the complexity and scalability issues mentioned above. New areas where startups are likely to enter and disrupt include teleoperation modules, controllers that allow for safer human-robot collaboration, human-centered AI protocols, and sensors that can quantify human behavior.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="c634">Eventually, we may see unified operating systems for AI-enabled robots similar to Windows or Android in the consumer electronics space, or standard user interfaces that will fundamentally transform the fragmented robotics industry and further accelerate its disruption.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="6144">Standard modules and interfaces will bring down the cost to enter the market and shorten the time to market. As DRL and robotics technology improves, we will see a boom across industries including agriculture, food processing, home, and even surgical robotics.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="2e46"><strong class="lj lv">How technology will transform the industry landscape</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="6586">Led by industry leaders like Amazon and Ocado, a British online grocery store, retailers see efficient and cost-effective supply chains as a key to success and are incentivized to automate their warehouses for lower cost, lower management complexity, and higher scalability. While Amazon keeps its Kiva mobile robots and automated warehouse solutions for in-house use only, Ocado launched its fulfillment platform to help other companies store, pick, and transport groceries. But most retailers still rely on material-handling companies and system integrators to build and maintain an automated warehouse.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="0f22">Incumbent material-handling companies tend to exploit the current technology, focusing on high-revenue-potential retrofit projects. This gives startups an opportunity to enter from the lower-end markets, where incumbents are less interested, and thus would not respond immediately. Dozens of startups offer scalable and affordable solutions that address customer needs for robots that can handle high-variability, low-precision tasks, such as piece picking in the warehouses.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="5539">On the other hand, large robotics companies have long focused on developing high-speed and more-precise solutions to serve their main customers in the automotive and manufacturing industries. The pursuit of absolute precision is contrary to the process of deep learning, where results improve over time through trial and error.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="7bbf">Some robotics companies saw an opportunity to enter new segments that require robots to handle high-variability, but lower-precision tasks, and began to build in-house AI teams and explore partnerships with startups developing deep neural network solutions for robots.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="c818">There are currently more than 20 warehouse piece-picking robot startups worldwide, including Osaro and Righthand Robotics. Most of these startups (between seed round and Series C funding) are working on pilot projects with incumbent material-handling, retail, and robotics companies.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="416d">Software and services will also capture more values, while premiums in hardware will decrease. With this power shift, as well as potential mergers and acquisitions, the landscape will change significantly over the next few years.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="4459"><strong class="lj lv">The future of AI-enabled robotics</strong></p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="4e2a">Thoughts vary on the effect that AI and DRL will have on the robotics industry. One scenario is the robotics industry will consolidate because of the AI network effect, which says results improve as the amount of training data they’re given increases. This idea is that whoever owns the most proprietary data will win the competition.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="aef7">However, new researches in the areas of Transfer Learning, Meta Learning, and Zero-Shot Learning make it possible to learn from fewer and fewer examples, generalizing learning results across different fields. Cloud-based AI and robotics will also help accelerate learning and diffusion.</p><p class="lh li dz ap lj b lk ll lm ln lo lp lq lr ls lt lu" data-selectable-paragraph="" id="8b8a">As these technologies evolve, the landscape may face yet another significant change. This all starts from the warehouse!</p>

IIoT

Latest Posts

Industry 4.0 Deep Dive. Part 5: Flexible manufacturing

SRAM PUF: The Secure Silicon Fingerprint

Industry 4.0 Deep Dive. Part 1: Introducing smart manufacturing

How To Find Leaks In Pneumatic Systems

7 Benefits of AI In Manufacturing

Why is Warehouse Automation So Important for AI Professionals?

Profiles