ParticleTensorAI – AI-Based Analysis of Spray and Particle Signals

ParticleTensorAI, or PTA for short, is an AI-based analysis technology for spray diagnostics, particle measurement, fluid mechanics, and aerodynamics.

The technology evaluates time-resolved light-scattering signals from droplets or particles. Instead of analyzing only individual signal peaks, ParticleTensorAI processes complete signal sequences. As a result, it can identify complex patterns that classical evaluation methods may not detect.

ParticleTensorAI can support scientific research, product development, process monitoring, and industrial quality control.

A ParticleTensorAI measurement system can include:

• an optical measurement probe,
• an electro-optical unit,
• a high-speed data acquisition system,
• ParticleTensorAI analysis software,
• trained AI models for the required measurement task.

The exact configuration depends on the application. Therefore, the number of detectors, sampling rate, measurement duration, and AI model can vary.

ParticleTensorAI evaluates time-resolved light-scattering signals generated by droplets or particles as they pass through an optical measurement volume.

An optical probe illuminates the measurement region. When droplets or particles cross the light beam, they scatter light. Fast optical detectors then record the resulting signals.

The underlying optical measurement setup can use the TSTOF measurement principle. However, ParticleTensorAI describes the AI-based data analysis technology and not the physical measurement principle itself.

ParticleTensorAI converts recorded signal sequences into multidimensional data structures called tensors.

For example, the system can arrange several detector signals in separate channels. It can then transform a defined time window into an image-like tensor. A neural network analyzes this tensor and searches for characteristic signal patterns.

This approach allows the software to evaluate:

• signal shape,
• signal intensity,
• time shifts between detectors,
• temporal signal sequences,
• particle or droplet event density,
• statistical patterns within the measurement window.

In addition, ParticleTensorAI can combine information from several detectors in one AI model.

Classical counting methods often require a trigger threshold to identify individual droplet events. However, this can become difficult when signals overlap or when the spray contains many droplets.

ParticleTensorAI can analyze complete buffered signal sequences without detecting every signal event separately first.

Therefore, the method can also process:

• overlapping light-scattering signals,
• dense droplet sequences,
• signals with changing amplitudes,
• complex spray structures,
• measurement windows with many individual events.

This trigger-free approach is one of the main differences between ParticleTensorAI and classical droplet-counting methods.

The available results depend on the optical setup, the recorded training data, and the selected AI model.

Possible applications include:

• spray classification,
• particle or droplet classification,
• detection of process changes,
• comparison of different spray conditions,
• analysis of droplet composition,
• estimation of refractive index,
• estimation of mass concentration,
• identification of different liquid or particle classes,
• monitoring of stable and unstable spray conditions.

In research studies, ParticleTensorAI has also been investigated for the analysis of complex droplets and high-concentration sprays.

ai-quanton develops and manufactures the optical measurement probes in-house.

We can adapt the probe geometry and optical configuration to the required measurement task. For example, the probes can differ in:

• working distance,
• measurement volume,
• number of detectors,
• detector arrangement,
• laser power,
• optical access,
• mechanical design,
• installation position.

As a result, ParticleTensorAI can work with different laboratory setups, test rigs, and industrial measurement environments.

ParticleTensorAI can analyze signals from one or several optical detectors.

Depending on the measurement task, a system may use:

• one detector for basic signal classification,
• two detectors for additional time-shift information,
• three or four detectors for more complex signal patterns.

Several detector channels can provide more information about the passage of droplets or particles through the measurement volume. However, the best configuration always depends on the application.

The electro-optical unit connects the measurement probe to the data acquisition system.

It contains the optical and electronic components required to generate, detect, amplify, and transfer the light-scattering signals.

Depending on the configuration, the unit can also provide interfaces for:

• autonomous measurement operation,
• external triggering,
• analog outputs,
• digital communication,
• network connection,
• integration into test rigs or production systems.

ParticleTensorAI can provide both raw measurement data and AI-generated results.

The available output may include:

• time-resolved detector signals,
• buffered signal sequences,
• tensor data,
• classification results,
• predicted process parameters,
• probability values,
• statistical measurement results,
• trend data,
• exportable result files,
• network-based process values.

Therefore, users can apply the data for direct monitoring, model development, scientific analysis, or process control.

SprayQuantAI and ParticleTensorAI use different evaluation approaches.

SprayQuantAI mainly focuses on identifying and evaluating individual droplet events. It can determine parameters such as droplet size, velocity, and droplet rate through a counting-based analysis.

ParticleTensorAI, in contrast, analyzes complete signal sequences as tensors. It does not always need to separate each individual droplet signal before the evaluation.

Therefore:

• SprayQuantAI is mainly a counting-based measurement and analysis approach.
• ParticleTensorAI is mainly a tensor-based and AI-driven signal analysis approach.

Both technologies can use optical light-scattering signals. However, they evaluate the recorded data in different ways.

Yes. ParticleTensorAI can be adapted to customer-specific measurement tasks.

Possible adaptations include:

• optical probe geometry,
• number of detectors,
• detector arrangement,
• sampling rate,
• buffer duration,
• tensor dimensions,
• AI model architecture,
• output parameters,
• data interfaces,
• integration into existing systems.

Before developing the final configuration, we evaluate the spray, particle system, measurement environment, available reference data, and required output parameters.

Yes. An AI model needs suitable training and validation data for the intended measurement task.

The quality of the results depends strongly on:

• the quality of the reference measurements,
• the range of conditions in the training dataset,
• the stability of the optical setup,
• the number of recorded measurement cases,
• the selected model architecture,
• the similarity between training and application conditions.

Therefore, customer-specific applications may require an individual measurement and training campaign.

ParticleTensorAI can support fast or near-real-time analysis, depending on the hardware, buffer size, AI model, and required output rate.

For industrial monitoring, the system can evaluate consecutive measurement windows and provide process values or classifications through digital or network interfaces.

However, the achievable update rate depends on the selected configuration and the complexity of the AI model.

DECHEMA 2026 in Frankfurt – ParticleTensorAI for AI-Assisted Spray and Particle Characterization

Several measurement, data-processing, and AI-based evaluation concepts associated with ParticleTensorAI are covered by patent applications and intellectual property rights.

The implementation also requires expertise in:

• optical measurement technology,
• light scattering,
• high-speed data acquisition,
• signal processing,
• machine learning,
• fluid mechanics,
• spray and particle diagnostics.

Yes. Scientific publications, technical reports, conference contributions, and patent applications describe the underlying optical measurement methods and AI-based evaluation concepts.

The available studies cover topics such as:

• AI-based evaluation of light-scattering signals,
• classification of individual droplets,
• analysis of buffered signal sequences,
• determination of refractive index,
• estimation of droplet composition,
• high-concentration spray measurement,
• tensor-based analysis of optical detector data.

A selection of publications and technical documents can be provided in the Publications and References section.

Technical descriptions, available datasheets, publications, and application examples can be provided in the download section of the website.

Because many ParticleTensorAI systems require a customer-specific configuration and AI model, the technical specifications may vary between applications.

ai-quanton develops the ParticleTensorAI technology, optical measurement probes, electronics, and analysis software.

Commercial consulting and sales can be handled through the sales partner AOM-Systems.

For a technical evaluation, customers should provide information about:

• the spray or particle system,
• the expected particle or droplet properties,
• the process conditions,
• the required measurement results,
• the available reference data,
• the planned system integration.

The price depends on the optical probe, number of detectors, data acquisition hardware, software configuration, AI model development, and required system integration.

Customer-specific AI training or validation measurements may also influence the project cost.

For an individual quotation, please contact the sales partner AOM-Systems.