We'll help you to build ML algorithms for your machines and devices from your data.
Custom Quotation - Contact Us
Predictive Maintenance, Time-to-Maintenance prediction of Consumables, Machine Programs Optimization, Advanced Diagnostics, Anomaly Detection, Private Federated Learning.
Edge AI Working Scheme
Edge AI represent AI models optimized to be installed on Edge Hardware (SoM, SoC, Microcontrollers, Edge CPUs) to perform real-time predictions diretcly at the place of data generation.
Advantages:
Real-Time Inference
Ability to process data and provide predictions in a very low fraction of time. Close to zero latency.
Hardware Optimized
Optimized to run on a specific hardware architecture
Low Costs
It can be integrated on existing harware. Optimized to reduce energy consumption and computing power. There is no mandatory need to transfer and store data.
Offline Operativity
Ability to operate without any internet connectivity because they are integrated into devices and machines and there is no need to trasfer data for inference.
Applications:
REAL TIME ANOMALY DETECTION
Solutions capable to identify anomalies in the data generated and potentially generate triggers for fast response tio events.
AUTONOMOUS MACHINE MANAGEMENT
Solutions that are capable to learn the best performing practices from past working data and become able to self-manage machine parametrization and operativity
EMPLOYEES SAFETY
Solutions capable to identify dangerous conditions and trigger alarms or machine stops.
REAL-TIME DIAGNOSTICS
Solutions capable to identify and categorize the motivations behind unplanned machine failures.
REAL-TIME QUALITY CONTROL
Solutions capable to identify defects in machines' output by exploiting computer vision, vibrational and sound diagnostics.
VIRTUAL SENSORS
Solutions capable to merge the signals of several physical sensors to extract new information about conditions or properties.
CRITICAL CONDITIONS DETECTION
Solutions capable to identify machine's critical operational conditions that may compromise machines and trigger alarms or machine stops.
SMART PROGRAMS
Solutions capable to learn the patterns of a certain desidered output and self-manage the machine parameters to guarrantee the expected results under every external condition.
Are techniques designed to optimize Algorithms an Neural Networks for specific hardware implementation. The main scope of this activity is reducing computational cost, energetic consumption and model size.
Examples:
Quantization
Is an approximation process that modifies bit numbers of model weights (eg. from 64 bit to 8 bit)
Pruning
Pruning is a technique that cuts the neurons with a low impact on the functioning of the network.
Apache TVM
It's a deep-learning compiler framework that empowers engineers to optimize and run computations efficiently, on any hardware backend.
Example of Apache TVM framework
Private Federated Learning Working Scheme
A new decentralized & collaborative learning approach that guarrantees data integrity and sensitive data protection by-design
Advantages:
Raw Data Protection
Raw data is not necessarily transferred from the local devices. This allows full control on intellectual property and personal data.
Reduction of Data Transfer Cost
This approach reduces data transfer by 95%. This strongly reduce the need of huge servers to store data and perform the training.
Decoupling Ownership
A simple way to define ownership. Data remains customer's owned, models are created and owned by you.
Bandwidth Reduction
Parameters sharing reduces the amoun of data transfer of 95%. This makes it the best option for AloT projects in conditions of low connectivity.
Collaborative Approach
Merge capabilities while preserving data integrity and ownership.
Applications:
DEFENSE
Protect sensitive raw data coming from battlefield while training AI algorithms in real operational conditions.
Examples:
HEALTHCARE
Preserve patients data and create strong disease detection algorithms to early detect pathologies.
Examples:
PUBLIC ADMINISTRATIONS
Solutions capable to improve PA operations without infringing people's data privacy.
Examples:
COO-PETITION PROJECTS
Cooperation between competitors to reach a common goal by enlarging the database without infringing company's data IP ownership.
Examples:
Virtual sensors are models that are capable to estimate product properties, processes or conditions by fusing data of several physical sensor readings.
Examples:
Pulmonary Disease Detection
Identification of pathology presence patterns on data from enviromental air quality sensors
Acquaplanning Detection
Early detection of acquaplanning by merging data from several car sensors.
Smart Barista
Identify patterns and replicate good practices of best tasting coffee making.
Radiators Freezing
Identify early patterns of HVAC radiators freezing by merging data from in-machine sensors.
Sensor Fusion Scheme
We don't want you will feel alone while using our product.
Customer Support
Things5 customer support team is ready to help you with all your technical questions, and can be reached by our slack channel, email, or phone depending on your SLA.
Premium Integration Support
Get up and running quickly with a premium integration support plan. Then maximize results along the way with premium training, ongoing consulting, and technical services.
Servitization Support
Things5 advisory team is ready to help your company in design, build and implement your company's servitization strategy. Stay aligned with market needs, modern business models and best practices.
We Build Advanced Capabilities for your Things
All Rights Reserved | Visup Srl | VAT: IT09299620964