EZAKO and XPERT join forces to develop Deep Learning for all underwater platforms.
[Sophia-Antipolis, 8th April 2025]
Underwater data collection systems and platforms are diverse: underwater drones, buoys, submerged hydrophones, sonars… Their technical capabilities also vary: operating depth, types of onboard sensors, computing power, transmission capacity… Moreover, each platform collects data in a different environment: geographical diversity, varied marine life, different acquisition depths…
To be generalizable, artificial intelligence algorithms must therefore, adapt to each carrier.
This technological development project aims to generalize the adaptation of neural networks across all underwater platforms. The two partners will work together to improve real-time detection of underwater anomalies. These new capabilities will address the growing needs of oceanographers and military forces for at-sea data collection and processing, with the goal of achieving greater seafloor transparency and better understanding of the underwater environment.
For example, thanks to the joint developments of the two companies, it will become more efficient to identify sounds of interest related to port approaches, characterize detected vessels, and enhance the protection of sensitive areas.
This project benefits from the RAPID program (Support Scheme for Dual Innovation), which funds dual-use innovation projects through the Defense Innovation Agency (AID). In this effort, the two expert companies, Ezako and Xpert, received support from the GIMNOTE naval innovation cluster. The RAPID project is technically overseen by experts from the French Directorate General of Armament – Naval Technologies, in collaboration with the French Navy.
[Paris, 07/06/2023] – Ezako, a leading provider of cutting-edge AI software solutions, is pleased to announce its strategic partnership with Altair, a global technology company. This collaboration will enable Altair to distribute Ezako’s revolutionary Upalgo Labeling software through the renowned Altair Partner Alliance, expanding the availability of this innovative product to a wider market.
With the integration of Upalgo Labeling into Altair’s portfolio, customers will gain access to a powerful tool designed to simplify and expedite the annotation of time series and sensor data. The intuitive user interface and advanced AI-powered label propagation feature of Upalgo Labeling offer Altair customers an unparalleled opportunity to enhance their operations by efficiently organizing and analyzing their time series and sensor data. Industries dealing with substantial volumes of data, including defense, automotive, manufacturing, and energy, will particularly benefit from the capabilities of this software.
Upalgo Labeling empowers users to unlock the true potential of their data, enabling them to forecast future trends, detect anomalies, and identify meaningful patterns and correlations.
By joining forces with Altair, Ezako reinforces its commitment to providing industry-leading software solutions that drive innovation and enable businesses to optimize their data-driven processes. Altair’s global presence and extensive customer base, which spans across more than 11,000 organizations, make it an ideal partner to bring the transformative benefits of Upalgo Labeling to the forefront.
As part of the SecurIT program, EZAKO and MION have partnered to develop an innovative solution to detect potentially dangerous and illegal substances in luggage and goods.
Your company is probably collecting huge amounts of data. A big part of them might come from sensors and thus in the form of time-series data. Have you ever wondered what could you do with all this data? How to add value? How to exploit them? What ROI can they bring to your company?
You are at the right place. Ezako’s purpose is to help companies leverage time-series data. This article’s intention is to be a beginner’s guide to time-series data. Let’s start.
What is time-series data?
A time-series is a sequence of data taken at successive, often equally spaced, points in time.
Here an example where we have 4 timestamps with 2 values collected for every time-point :
00:00:01 12 0.2
00:00:02 13 0.1
00:00:03 16 0.3
00:00:04 14 0.4
Those values are time-series.
Where can we find these time-series data?
Time-series are found in more places than meets the eye. Some time-series are obvious:
Monitoring the evolution of temperature throughout the day is an example of a time-series. Recording the temperature is achieved through the use of a temperature sensor. For example: Thermocouple, thermistor and a thermostat.
The price of a stock is a time-series. The stock price evolves through time and is set on markets through a demand and supply ratio.
Time-series are also found in the industrial world. At an industrial scale, one particular plant may use thousands of different sensors of many different kinds, all generating time-series data continuously. In fact, most time-series around the world are collected from sensor data. Here are a few examples:
Temperature sensor
Proximity Sensor
Accelerometer
IR Sensor
Pressure Sensor
Light Sensor
Ultrasonic Sensor
Smoke, Gas and Alcohol Sensor
Touch Sensor
Humidity Sensor
Flow and Level Sensor
In 2020, more than 250 billion sensors were estimated to be in use; this number will grow to 1 trillion in 2025!
The vast majority of these sensors collect time-series data which can be used to drive value in a variety of different use cases: monitoring the health of equipment, testing prototypes and building prediction models…
How can you leverage Machine Learning in time-series data?
The huge amount of time-series data available makes it impossible to be analyzed without innovative machine learning techniques. But once mastered, these applications can be very powerful :
Anomaly Detection. Anomaly detection, also called outlier detection, is the identification of unexpected events, observations, or items that differ significantly from the norm. Typical algorithms used for anomaly detection are: isolation forest, one-class SVM, LSTM-autoencoder, ARIMA etc…
Prediction. Predictive analytics is the use of historical data in order to predict future events. Typical algorithms we can use for this are naive bayes, generalized linear models, logistic regression etc…
Other types of ML, such as classification, pattern recognition and labeling.
So how does this bring an ROI to me?
There are several applications possible. Each bringing more or less a subsequent ROI. Every day new use cases are tested, approved and deployed. Here are a few use cases where ML applied to your time-series data will bring a real ROI to your company :
Regression testing of complex hardware. Engineers spend 30% of their time testing their hardware after firmware and middleware updates. Algorithms help optimize this. By learning on stable versions and detecting anomalies on the new versions, machine learning algorithms can be 10x faster and 3x more accurate in detecting regressions than human teams.
Asset monitoring. If you have a fleet of expensive and complex assets, you would like to monitor them. Actual methods would often use static thresholds for alerting. Actual state of the art in ML is much more evolved. Today ML tools can monitor your assets for alerting in case of known problems but also Zero-day faults that have never been seen. This enables you to act fast on future problems and resolve them.
Supply chain optimization. Product promotions and punctual high demands or manufacturing problems are the usual problems that supply chain teams have to deal with. Finding anomalies before they happen and predicting demand beforehand, data science teams can help supply chain teams to be more efficient in their capacity planning, continuous activity but also storage optimisation.
Pricing optimization. Predicting demand by taking into account the seasonality and competitor pricing, can help companies to optimize their product pricing, thus maximising profitability.
And more…
If you want to talk about your own time-series data and how to leverage them, please send us an email: contact@ezako.com
Ezako has today been accepted to the Minc Scale program. This program is designed to support growing companies with the ambition to go international.
Ezako is specialized in the field of artificial intelligence, more precisely in machine learning. Its main activity is the analysis of time-series data. Ezako has three applications: pattern recognition, anomaly detection, and time-series labeling.
With Minc scale, Ezako will benefit from many advantages. The program will allow Ezako to penetrate the Nordic market with the help of a large community of coaches and mentors who support start-ups in their growth. Through this program, Ezako will be able to develop new skills with the help of documentation, tools and webinars offered.The programme will allow ,the expansion of its network and get more business partners. Minc scale also offers help with financing, marketing and sales.
Minc scale aims to bring together different start-ups in their growth phase and to share experiences and skills. This program accompanies start-ups with various supports such as marketing, expert advice, and sales.
Sophia Antipolis, France—May 05, 2022—Ezako today announced it has joined NVIDIA Inception, a program designed to nurture startups revolutionizing industries with technology advancements.
EZAKO is focused on artificial intelligence more precisely on machine learning. The main activity is the analysis of time-series data. Ezako has three applications: pattern recognition, anomaly detection, and time-series labeling.
NVIDIA Inception will allow Ezako to gain more visibility but also give support regarding some marketing activities. It will provide leading-edge resources in artificial intelligence, data science and also access to a large network of data scientists and experts. With NVIDIA Inception, Ezako will build a better network and unlock new commercial partnerships. The program will also offer Ezako the opportunity to collaborate with industry-leading experts and other AI-driven organizations.
“With NVIDIA Inception, we hope to extend our network and gain more knowledge on AI-related sectors such as deep learning, and machine learning. NVIDIA Inception will provide us, with technical support that can lead to great results for the growth of Ezako” – Bora Kizil, managing director.
NVIDIA Inception helps startups during critical stages of product development, prototyping, and deployment. Every NVIDIA Inception member gets a custom set of ongoing benefits, such as NVIDIA Deep Learning Institute credits, marketing support, and technical assistance, which provides startups with the fundamental tools to help them grow.
About Ezako,
Ezako is an innovative company based in Paris and Sophia-Antipolis. Ezako’s goal is to help its customers in the automatic detection of anomalies.
Ezako proposes a software solution called UPALGO, which allows the labeling of data in a robust and fast way, and produces specifically supervised, unsupervised and multivariate anomaly detection algorithms. The possible applications are numerous: prediction of operational problems, prediction of system failures, monitoring of assets, and IoT. Thanks to this Deep Learning technology and the team of experts, complex testing, validation, and monitoring operations become more efficient and less costly.
The meantime between failures, also called MTBF, is the average time between system breakdowns. It can be used as a reliability indicator. MTBF can also indicate if a machine has been badly repaired or badly used by human beings.
Why is it important?
A machine that breaks down disables the whole organization. When the company’s operations depend on these machines, their downtime represents a considerable cost.
This is why performance measures are essential for the proper functioning of the company. Measuring MTBF is the first step in the goal to minimize downtime.
How to calculate?
To calculate the mean time between failures, you have to divide the total number of operational hours in a period by the number of failures that occurred in that period. It is usually measured in hours.
MTBF = Number of operational hours ÷ Number of failures
For example, let’s say you have a piece of equipment that is supposed to work for 12 hours. In the end, it worked for 10 hours instead. It broke down 4 times for a total duration of 2 hours.
MTBF = 10÷4
MTBF = 2,5 hours
Here, we have calculated that the MTBF is 2,5 hours. It is usually measured in hours.
How to decrease it?
You can combine MTBF with other maintenance calculators like the meantime to failure (MTTF) in order to avoid a breakdown. In fact, it will help the maintenance team to improve their system and be more efficient. The checking of assets can be done through the validation process, testing process, health monitoring and failure prediction etc…
All this process, calculating MTBF and an anomaly detection tool can make you save money and work faster.
Federated learning is a method of training machine learning models on decentralized datasets, where the data is distributed across multiple devices, such as smartphones or computers, in a decentralized manner. This approach allows for training more accurate models by leveraging the data available on a large number of devices, without the need to centralize the data in a single location. This can be especially useful in situations where the data is sensitive and cannot be centrally collected and stored. In federated learning, the participating devices train a shared model by sending their local updates to a central server, which aggregates the updates to improve the global model. This process is repeated iteratively until the model has converged. [ChatGPT]
A major challenge related to engine development testing is to detect abnormal observations in high-frequency sensor time series, which can hinder their convenient analysis. In practice, the use of these signals involves the calculation of spectrograms and the tracking of the vibration amplitude of different physical phenomena. The presence of these abnormal observations in the signals can go as far as to make these spectrograms completely unusable. The detection and correction of these abnormal observations become essential to continue to use the information contained in these sensors.
For proper signal analysis, an industrial anomaly detection technique has been developed by Safran Aircraft Engines. Even though this approach yields good results, it suffers from an important drawback which is the extensive intervention of an expert via an incremental fine-tuning of parameters.
To circumvent this limitation, Ezako proposed two alternative strategies, one being a variant of a classical statistical technique developed by EZAKO named Rolling Standard Deviation and the other being an architecture of the well-celebrated Deep Learning method named LSTM (Long Short Term Memory).
The study shows how our work resulted in very good detection performance while being up to 10 times faster than the expert-based method.
To download our published joint paper, please click here: Download the paper.
DETECTION OF ANOMALIES ON SIGNALS DURING AIRCRAFT ENGINE TESTS: METHODOLOGICAL COMPARISON BETWEEN HISTORICAL, STATISTICAL AND DEEP LEARNING APPROACHES. (Yann Rotrou (SAFRAN), Mourad YAHIA BACHA (SAFRAN), Julien MULLER (EZAKO), Yacine EL AMRAOUI (EZAKO))
Anomaly detection is the process of identifying unusual patterns in data that do not conform to expected behavior. This can be incredibly important for companies, as it allows them to identify potential problems or abnormalities in their operations before they become serious issues.
One of the key reasons that anomaly detection is so important for companies is that it allows them to identify potential issues in real-time. By detecting anomalies as they happen, companies can take immediate action to address the problem before it becomes a major issue. This can help to prevent costly downtime, avoid lost revenue, and protect the company’s reputation.
Another reason that anomaly detection is important for companies is that it can help to improve efficiency and productivity. By identifying and addressing abnormalities in operations, companies can optimize their processes and ensure that they are running smoothly. This can help to reduce waste and inefficiency, leading to cost savings and improved performance.
Anomaly detection can also help to improve the accuracy and reliability of a company’s data. By identifying and addressing abnormalities in data, companies can ensure that their data is accurate and trustworthy. This is particularly important for companies that rely on data to make important decisions, as accurate data is essential for making informed choices.
In addition to the benefits mentioned above, anomaly detection can also help to improve the security of a company’s operations. By detecting unusual patterns in data, companies can identify potential security threats and take action to protect their assets and sensitive information. This can help to prevent data breaches and other security incidents, protecting the company and its customers.
Overall, anomaly detection is an important tool for companies, as it allows them to identify potential problems and abnormalities in their operations in real-time. By detecting and addressing these issues, companies can improve their efficiency, accuracy, and security, leading to better performance and a stronger bottom line.
