https://neurocontrols.io/en/blog/Engineering articles on multimodal sensing, condition monitoring and edge AI.enhttps://neurocontrols.io/en/blog/electronica-2026-neurocontrols-live-demo/https://neurocontrols.io/en/blog/electronica-2026-neurocontrols-live-demo/Tue, 13 Oct 2026 09:00:00 +0000Clemens SaurProducts & newsNeuroControls exhibits at electronica 2026 at Messe München from 10 to 13 November 2026, in Hall B4 at stand 169/4. Visitors can see the multimodal sensing platform live at our stand, with every stream time-stamped on the hub. Meetings with our engineers can be booked in advance through the contact page.https://neurocontrols.io/en/blog/infrared-impulse-thermography-subsurface-defects/https://neurocontrols.io/en/blog/infrared-impulse-thermography-subsurface-defects/Sat, 03 Oct 2026 09:00:00 +0000Clemens SaurMultimodal sensingInfrared impulse thermography detects sub-surface defects by heating a surface with a short pulse of light and recording how it cools with an infrared camera. Voids, delaminations and poor bonds slow the flow of heat into the part, so the surface above them stays warmer for longer. The time at which a defect appears grows roughly with the square of its depth.https://neurocontrols.io/en/blog/from-research-rig-to-product/https://neurocontrols.io/en/blog/from-research-rig-to-product/Fri, 25 Sep 2026 09:00:00 +0000Clemens SaurProducts & newsWe develop custom sensing systems in stages. A feasibility study tests whether a physical signal carries the information at all. A research rig records many channels at high rates to find out which ones matter. A demonstrator proves the reduced set in a relevant environment. Only then is the system hardened into a product with self-monitoring sensors, reliable recording and models trained on real data.https://neurocontrols.io/en/blog/iso-17359-iso-13374-condition-monitoring-standards/https://neurocontrols.io/en/blog/iso-17359-iso-13374-condition-monitoring-standards/Fri, 18 Sep 2026 09:00:00 +0000Saichand GourishettiStandards & regulationISO 17359 and ISO 13374 are complementary condition monitoring standards from ISO/TC 108/SC 5. ISO 17359:2018 gives the general procedure for setting up a monitoring programme, from cost-benefit analysis and criticality audit to monitoring method, alarm criteria and review. ISO 13374 defines how monitoring data is processed, communicated and presented, in six blocks from data acquisition to advisory generation.https://neurocontrols.io/en/blog/iso-20816-vibration-severity-zones/https://neurocontrols.io/en/blog/iso-20816-vibration-severity-zones/Fri, 11 Sep 2026 09:00:00 +0000Saichand GourishettiStandards & regulationISO 20816 vibration severity zones classify a machine's broadband vibration into four bands. Zone A is typical of newly commissioned machines, zone B is acceptable for unrestricted long-term operation, zone C is unsatisfactory for long-term continuous operation, and zone D is severe enough to cause damage. The boundaries depend on machine type, size and support, and they are guidance, not acceptance specifications.https://neurocontrols.io/en/blog/neuromorphic-computing-condition-monitoring/https://neurocontrols.io/en/blog/neuromorphic-computing-condition-monitoring/Thu, 03 Sep 2026 09:00:00 +0000Clemens SaurEdge AI & dataNeuromorphic computing for condition monitoring means processing sensor signals with event-driven hardware and spiking neural networks that only compute when the input changes. Vibration or sound is converted into spikes, a small network detects deviations, and the chip idles between events. Commercial neuromorphic processors now run audio and radar tasks below one milliwatt, which makes them candidates for battery-free sensors, though a well-designed microcontroller pipeline can still compete.https://neurocontrols.io/en/blog/digibatt360-battery-passport-data-space/https://neurocontrols.io/en/blog/digibatt360-battery-passport-data-space/Thu, 27 Aug 2026 09:00:00 +0000Clemens SaurBattery production & weldingDIGIBATT360 is a German research project, running from February 2026 to January 2029, that extends the EU digital battery passport with a federated data space and AI-based analytics. Led by VARTA Storage with Fraunhofer IIS, Circulor, VRI and NeuroControls, it aims to make lifecycle data for LMT and industrial batteries available across companies, to support state-of-health forecasts, second-life decisions and recycling.https://neurocontrols.io/en/blog/edgeai-vcd43-energy-autonomous-freight-sensor/https://neurocontrols.io/en/blog/edgeai-vcd43-energy-autonomous-freight-sensor/Thu, 20 Aug 2026 09:00:00 +0000Clemens SaurRail & energy-autonomous sensingVCD4.3 is an energy-autonomous sensor for freight wagons developed in the EU EdgeAI project. A piezoelectric harvester bolted to the bearing housing turns wheel impacts into power, a low-power microcontroller grades wheel flats and bearing faults with time-domain features in under 10 µJ per inference, and the result is written into a dual-interface RFID tag that trackside readers collect. No battery, no radio, no raw data leaves the wagon.https://neurocontrols.io/en/blog/industrial-data-ownership-eu-data-act/https://neurocontrols.io/en/blog/industrial-data-ownership-eu-data-act/Wed, 12 Aug 2026 09:00:00 +0000Saichand GourishettiStandards & regulationUnder the EU Data Act, Regulation (EU) 2023/2854, machine data is not "owned" in a property sense. The user of a connected machine gets the right to access the raw and pre-processed data it generates and to share it with third parties. The data holder, often the machine builder, may use non-personal data only under a contract with the user. The Act has applied since 12 September 2025.https://neurocontrols.io/en/blog/weld-traceability-sensor-to-module-record/https://neurocontrols.io/en/blog/weld-traceability-sensor-to-module-record/Wed, 05 Aug 2026 09:00:00 +0000Saichand GourishettiBattery production & weldingPer-weld traceability means every weld in a battery module can be traced from its quality result back to the module identifier, the joint position, the raw sensor data, the time stamp, the station and the model version that graded it. It requires consistent identifiers, time stamps with known uncertainty, integrity protection such as hashes, and open export formats, so that the record holds up years later.https://neurocontrols.io/en/blog/vibro-acoustic-sensing-medical-research/https://neurocontrols.io/en/blog/vibro-acoustic-sensing-medical-research/Wed, 29 Jul 2026 09:00:00 +0000Saichand GourishettiCondition monitoringVibro-acoustic sensing in medical research records the sound and vibration that the body produces, for example breath sounds at the chest wall, with microphones or contact sensors. The engineering problems mirror industrial monitoring: coupling between sensor and surface, background noise, small and inconsistently labelled datasets, and validation. A research prototype without a medical intended purpose is not a medical device; given a medical intended purpose it can qualify as one, and the EU Medical Device Regulation then governs its clinical investigation and placing on the market.https://neurocontrols.io/en/blog/digital-product-passport-espr-beyond-batteries/https://neurocontrols.io/en/blog/digital-product-passport-espr-beyond-batteries/Tue, 21 Jul 2026 09:00:00 +0000Clemens SaurStandards & regulationThe digital product passport under ESPR is a product-specific digital record, linked to a unique identifier on the product, that carries compliance, sustainability and safety information. Regulation (EU) 2024/1781 makes it mandatory for every product group that receives ecodesign rules, with steel, aluminium, textiles, tyres, furniture and mattresses prioritised. Batteries come first through their own regulation from 18 February 2027, but ESPR extends the idea across the economy.https://neurocontrols.io/en/blog/self-monitoring-sensors-false-alarms/https://neurocontrols.io/en/blog/self-monitoring-sensors-false-alarms/Tue, 14 Jul 2026 09:00:00 +0000Saichand GourishettiEdge AI & dataSelf-monitoring sensors continuously check the quality of their own signal, for example bias voltage, noise floor, clipping, dropouts and plausibility against other sensors. When a check fails, they raise a sensor alert and mark the affected data, so a loose mounting, damaged cable or saturated input is reported as a sensor problem instead of reaching the fault models as a machine fault.https://neurocontrols.io/en/blog/edge-vs-cloud-industrial-sensing/https://neurocontrols.io/en/blog/edge-vs-cloud-industrial-sensing/Tue, 07 Jul 2026 09:00:00 +0000Saichand GourishettiEdge AI & dataFor industrial sensing, evaluate high-rate sensor data at the edge, close to the machine, and send results, features and selected raw segments to the cloud. One vibration channel at 200 kSPS, stored at 4 bytes per sample, produces about 69.1 GB a day, or 6.4 Mbit/s of continuous uplink for that channel alone, and a machine usually has several such streams. The cloud remains the right place for fleet analytics, long-term storage and training models across sites.https://neurocontrols.io/en/blog/e-las-plus-results-multimodal-weld-inspection/https://neurocontrols.io/en/blog/e-las-plus-results-multimodal-weld-inspection/Mon, 29 Jun 2026 09:00:00 +0000Clemens SaurBattery production & weldinge-LAS+ was a research project with Fraunhofer IDMT, NeuroControls, SSI Software Service, SITEC and VRI as partners, funded by the German Federal Ministry for Economic Affairs and Climate Action, on machine-learning-based multimodal inline inspection of laser weld seams in battery cell production. Its project period was May 2023 to April 2026, according to the Fraunhofer IDMT project page, and its results were to be demonstrated in a pilot plant. For NeuroControls, the main outcome is the measurement chain behind our Process Monitoring Kit.https://neurocontrols.io/en/blog/multimodal-sensing-glossary/https://neurocontrols.io/en/blog/multimodal-sensing-glossary/Mon, 22 Jun 2026 09:00:00 +0000Saichand GourishettiMultimodal sensingThis multimodal sensing glossary defines 40 terms used when vision, thermal, depth, sound and vibration sensors are combined on industrial machines. It covers sensor types, signal acquisition, time-stamping, data recording, models and condition monitoring. Each definition is short and self-contained, so mixed teams of mechanical engineers, electronics designers and data scientists can agree on what words mean before they disagree about results.https://neurocontrols.io/en/blog/cnc-spindle-monitoring-tool-wear-chatter/https://neurocontrols.io/en/blog/cnc-spindle-monitoring-tool-wear-chatter/Mon, 15 Jun 2026 09:00:00 +0000Saichand GourishettiCondition monitoringCNC spindle monitoring detects tool wear and chatter by watching vibration, sound and spindle load during each cut. Chatter is a self-excited vibration that appears close to a natural mode of the tool or spindle, at frequencies that usually do not line up with the tooth passing frequency or, with runout, with any spindle speed harmonic. Tool wear shows as a slow rise in cutting load and vibration for the same operation, compared against earlier cuts.https://neurocontrols.io/en/blog/cyber-resilience-act-connected-sensors/https://neurocontrols.io/en/blog/cyber-resilience-act-connected-sensors/Sun, 07 Jun 2026 09:00:00 +0000Clemens SaurStandards & regulationThe Cyber Resilience Act, Regulation (EU) 2024/2847, sets cybersecurity requirements for products with digital elements, which includes connected industrial sensors, edge hubs and their software. Most sensors fall into the default category and can be self-assessed. Since 11 September 2026, manufacturers must report actively exploited vulnerabilities and severe incidents; from 11 December 2027, products must meet the essential requirements and carry the CE marking on that basis.https://neurocontrols.io/en/blog/edge-ai-hardware-qcs6490-npu/https://neurocontrols.io/en/blog/edge-ai-hardware-qcs6490-npu/Sun, 31 May 2026 09:00:00 +0000Clemens SaurEdge AI & dataA QCS6490-class hub is an edge AI computer built around Qualcomm's QCS6490 system-on-chip: an eight-core Kryo 670 CPU, an Adreno 643 GPU, a Spectra 570L image signal processor and a Hexagon 770 AI engine rated at 12 dense TOPS. For sensing, it can capture several cameras and other sensors, extract features and run models at the machine; the SoC itself is specified at a typical 6 to 9 W, and a complete hub draws more depending on its peripherals.https://neurocontrols.io/en/blog/acoustic-monitoring-laser-welding/https://neurocontrols.io/en/blog/acoustic-monitoring-laser-welding/Sun, 24 May 2026 09:00:00 +0000Saichand GourishettiBattery production & weldingAcoustic monitoring of laser welding records the sound emitted by the keyhole and melt pool, either through the air with a microphone or through the part with an acoustic emission sensor. The signal carries information on keyhole stability, energy input, part-to-part gap and events such as pores or blowouts. Machine learning models trained on destructively verified welds can grade each weld inline from this signal.https://neurocontrols.io/en/blog/passive-rfid-sensor-readout/https://neurocontrols.io/en/blog/passive-rfid-sensor-readout/Sat, 16 May 2026 09:00:00 +0000Clemens SaurRail & energy-autonomous sensingPassive RFID read-out lets a sensor deliver results without a radio or a battery. The sensor writes a few bytes into a dual-interface RFID tag over a wired bus such as I²C. When a reader passes, its field powers the tag, which replies by backscatter or load modulation. The reader supplies the communication energy, so the sensor spends almost nothing on transmitting data.https://neurocontrols.io/en/blog/labelling-multimodal-industrial-datasets/https://neurocontrols.io/en/blog/labelling-multimodal-industrial-datasets/Sat, 09 May 2026 09:00:00 +0000Saichand GourishettiMultimodal sensingLabelling multimodal industrial datasets works best when you design the label scheme before labelling, label each event in the modality where it is clearest, and propagate labels to the other streams using recorded time stamps and offset uncertainty. Use process records such as test results and maintenance logs as weak labels, double-label a sample to measure agreement, and document the dataset.https://neurocontrols.io/en/blog/hub-and-node-architecture-industrial-sensing/https://neurocontrols.io/en/blog/hub-and-node-architecture-industrial-sensing/Sat, 02 May 2026 09:00:00 +0000Clemens SaurProducts & newsA hub-and-node architecture splits a sensing system into one central hub, which records, time-stamps and presents data, and several nodes, each dedicated to one sensing function such as vibration, sound, vision or depth. We build one node per function because each physical quantity needs its own front end, short analogue paths and independent failure handling, and because functions can then be added or upgraded one at a time.https://neurocontrols.io/en/blog/predictive-maintenance-roi-pumps-motors/https://neurocontrols.io/en/blog/predictive-maintenance-roi-pumps-motors/Fri, 24 Apr 2026 09:00:00 +0000Saichand GourishettiCondition monitoringPredictive maintenance ROI for pumps and motors depends on four plant-specific numbers: how often assets fail unexpectedly, what an unplanned failure costs compared with a planned repair, what share of failures monitoring catches early, and how many false alarms it raises. Multiply avoided cost by caught failures, subtract false-alarm cost, and you have the yearly budget monitoring must beat.https://neurocontrols.io/en/blog/training-data-acoustic-quality-inspection/https://neurocontrols.io/en/blog/training-data-acoustic-quality-inspection/Fri, 17 Apr 2026 09:00:00 +0000Saichand GourishettiEdge AI & dataThere is no fixed amount of training data that acoustic quality inspection needs. The requirement depends on the task (anomaly detection needs only normal data; classification needs examples of every defect), on how clearly defects change the sound, and on how much normal production varies. The reliable way to size it is a pilot recording and a learning curve on grouped data.https://neurocontrols.io/en/blog/inline-vs-end-of-line-inspection-battery-modules/https://neurocontrols.io/en/blog/inline-vs-end-of-line-inspection-battery-modules/Fri, 10 Apr 2026 09:00:00 +0000Clemens SaurBattery production & weldingInline inspection checks each process step, such as every cell connection weld, while the module is being built. End-of-line inspection tests the finished module: voltages, insulation, communication, leak tightness. End-of-line tests prove the module works when it leaves the line; inline inspection catches defects that still pass those tests, such as mechanically weak welds, and finds them while rework is still possible.https://neurocontrols.io/en/blog/multimodal-test-rig-sensor-checklist/https://neurocontrols.io/en/blog/multimodal-test-rig-sensor-checklist/Thu, 02 Apr 2026 09:00:00 +0000Clemens SaurMultimodal sensingTo choose sensors for a multimodal test rig, start from the effects you must observe and their highest frequencies, sizes and durations. Then pick, per channel, a sensor whose bandwidth, resolution and frame rate cover them with margin, a sample rate of roughly 2.5 to 4 times the highest frequency of interest, a mounting that preserves the response, and a plan for time stamps and storage.https://neurocontrols.io/en/blog/eu-machinery-regulation-2023-1230-ai/https://neurocontrols.io/en/blog/eu-machinery-regulation-2023-1230-ai/Thu, 26 Mar 2026 09:00:00 +0000Clemens SaurStandards & regulationThe EU Machinery Regulation 2023/1230 replaces the Machinery Directive 2006/42/EC and applies from 20 January 2027. For AI and sensors, it adds three things: third-party conformity assessment for safety components and embedded systems with self-evolving machine-learning behaviour, essential requirements on protection against corruption and on logging software interventions, and specific rules for control systems that learn. Since the Digital Omnibus on AI of July 2026, the AI Act's requirements for machine safety components will also arrive through this Regulation. Advisory monitoring that performs no safety function is largely unaffected.https://neurocontrols.io/en/blog/airborne-ultrasound-leak-friction-detection/https://neurocontrols.io/en/blog/airborne-ultrasound-leak-friction-detection/Thu, 19 Mar 2026 09:00:00 +0000Clemens SaurCondition monitoringAirborne ultrasound leak detection works because gas escaping through a small opening becomes turbulent and emits broadband sound well above 20 kHz, the upper limit of human hearing. Friction in under-lubricated contacts does the same. Ultrasound is short-wavelength and attenuates quickly in air, so a sensor pointed at the source hears it clearly above the low-frequency noise of a factory and can locate it.https://neurocontrols.io/en/blog/validating-industrial-ai-shortcut-learning/https://neurocontrols.io/en/blog/validating-industrial-ai-shortcut-learning/Wed, 11 Mar 2026 09:00:00 +0000Saichand GourishettiEdge AI & dataValidating industrial AI models means testing them on data that differs from the training data the way real deployment will: other machines, other batches, other days. Shortcut learning happens when a model learns a cue that correlates with the label only in your data set, such as the recording session or station. Grouped splits and targeted probes expose it before deployment.