Systems and methods for enhanced particle detectors

Machine learning integration in optical particle counters addresses the challenge of detecting small particles and reducing noise interference, enhancing detection accuracy and reliability in cleanroom environments.

WO2026235952A1PCT designated stage Publication Date: 2026-11-12PARTICLE MEASURING SYSTEMS INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/US2026/025499
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-06
Filing Date
2026-04-28
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing optical particle counters struggle with accurately detecting small particles below 20 nm in size, particularly in cleanroom environments, and suffer from high false positive detection events due to noise interference, requiring frequent calibration and maintenance.

Method used

Integration of machine learning algorithms into optical particle counters for real-time noise reduction, adaptive calibration, and biological speciation, using supervised and unsupervised learning models to enhance detection accuracy and reliability.

Benefits of technology

Improves detection sensitivity and reduces false positives, allowing for accurate identification and differentiation of particles, especially biological contaminants, with reduced calibration needs and enhanced performance in diverse applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US2026025499_12112026_PF_FP_ABST
    Figure US2026025499_12112026_PF_FP_ABST
Patent Text Reader

Abstract

A method of detecting particles includes, passing a particle-containing fluid through a detection region of an optical particle detector; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles in the fluid and the beam results in scattered, emitted or transmitted electromagnetic radiation from the particles; directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector; generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector; processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output; and determining a particle measurement output, wherein the particle measurement output is based at least in part on the artificial intelligence-enabled particle counter output.
Need to check novelty before this filing date? Find Prior Art

Description

Leydig Ref. 340550: 29-25 WOSYSTEMS AND METHODS FOR ENHANCED PARTICLE DETECTORS CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Patent Application No. 63 / 800,750, filed May 6, 2025, which is hereby incorporated by reference in its entirety.BACKGROUND OF INVENTION

[0002] Advancement of technologies requiring cleanroom conditions has resulted in a need for the detection and characterization of smaller and smaller particles. For example, some industries pursue detection of particles less than 20 nm in size, and in some cases less than 10 nm in size, or even smaller, as they may affect the increasingly sensitive manufacturing processes and products. Similarly, the need for aseptic processing conditions for manufacturing of, e.g., pharmaceuticals and biomaterials requires accurate characterization of viable and non-viable particles to address compliance standard relating to health and human safety. Furthermore, accuracy in particle detection in manufacturing facilities is of the utmost importance, as production may be halted in response to a purported detection of contaminant particles.

[0003] Thus, numerous important and growing industries rely on optical particle counters for detection and characterization of small particles. The ability to detect smaller particles requires new approaches for optical particle counting such as systems employing increasing laser powers, shorter excitation wavelengths and more complex techniques to separate instrument noise from valid detection signals. These new approaches can also require more frequent calibration and maintenance to provide the necessary reliability and reproducibility. Furthermore, fast and accurate detection of particulate contamination, in concert with effective suppression of false positive detection events, is necessary in order to maximize production line uptime.

[0004] It can be seen from the foregoing that there is a need in the art for systems and methods that provide enhanced detection of particulate contamination, as wellLeydig Ref. 340550: 29-25 WOas improved functioning of the devices and user interfaces associated with particle monitoring equipment.SUMMARY OF THE INVENTION

[0005] Provided herein are optical particle counters that include integral machine learning functionality implemented on a real-time or near real-time basis in order to improve optical particle detector output and support in particle monitoring capabilities. For example, machine learning may be integrated into a data filter upstream of the particle data output to the user to improve particle detector output. Machine learning may also be integrated to support particle monitoring capabilities by, for example, enhancing sensitivity or lowering particle size thresholds.

[0006] The integration of machine learning techniques into optical particle counting systems may help to solve the above-mentioned shortcomings. For example, integrating machine learning techniques can improve calibration, noise reduction, biological speciation, anomaly detection, and adaptive measurements. By employing advanced algorithms, for example, supervised and unsupervised learning models, the optical particle counting system can adapt to varying environmental conditions and particle characteristics. Machine learning models can be trained on large datasets, allowing for accurate identification and differentiation of biological particles from nonbiological particles thus enhancing biological speciation.Additionally, these algorithms can detect anomalies in real time, facilitating immediate corrective actions and improving overall measurement accuracy. Through adaptive algorithms, the optical particle counter system can continuously refine its calibration and noise reduction processes, leading to enhanced performance and reliability across diverse applications.

[0007] Without wishing to be bound by any particular theory, there may be discussion herein of beliefs or understandings of underlying principles relating to the devices and methods disclosed herein. It is recognized that regardless of the ultimate correctness of any mechanistic explanation or hypothesis, an embodiment of the invention can nonetheless be operative and useful.

[0008] Various aspects of the present invention relates to machine learning. In particular, the model may be a machine learning model.Leydig Ref. 340550: 29-25 WO

[0009] Machine learning is a branch of artificial intelligence that involves the development of algorithms and models that allow computers to learn and make predictions or decisions without being explicitly programmed. It focuses on creating systems that can improve their performance over time by learning from data.

[0010] Training a machine-learning model refers to the process of teaching the model to make accurate predictions or decisions. During training, the model is exposed to a large amount of data, which is used to adjust the model's internal parameters or weights. The model learns patterns, relationships, or rules from the training data, allowing it to generalize and make predictions on new, unseen data.

[0011] Training data is the set of examples or instances that is used to teach a machine-learning model. It is often labeled data, meaning that each example is associated with a known outcome or target value. The training data consists of both input features and the corresponding output or target variable. The model learns from this data by analyzing the patterns and relationships between the input features and the target variable. Training algorithms, such as supervised learning, semisupervised learning, unsupervised learning or reinforcement learning may be used for training the machine-learning model.

[0012] Machine-learning models, such as the machine-learning model being trained in the present disclosure, are often implemented as Artificial Neural Networks (ANNs), and in particular Deep Neural Networks, Support Vector Machines, Decision Tree models, or Random Forest models.

[0013] The integration of machine learning may provide several advantages to optical particle counters including improved accuracy, especially in challenging environments or with low-cost sensors, enhanced classification of different types of particles, processing of data in real-time providing timely information for monitoring and decision-making, and making low-cost optical particle counters more accessible and practical in various applications.

[0014] In general, an artificial intelligence I machine learning model system includes inputs, processing steps, and outputs. In the case of the AI / ML system residing within a particle counter, the inputs typically come from the device hostingLeydig Ref. 340550: 29-25 WOthe system and may include data associated with the status or function of the laser, photodetector, fluid flow, fluid pressure, environmental measurements of variables such as temperature, relative humidity, vibration, etc. The inputs to the machine learning model may result in an improved / optimized data product output related to contamination measurement or outputs which result in modifications to the laser (ex. power level), photodetector (ex. gain or bias voltage) or fluid flow characteristics -subsequently also leading to an optimized contamination measurement result.

[0015] In the case where the AI / ML system resides external to the particle counter, the inputs may be the same as the case where the system is internal to the particle counter, but the inputs may be from many distinct particle counters or particle counter components. In some embodiments, the output may include optimized algorithms to improve the quality of contamination measurements in distributed particle counters or focus on predictive analysis of particle counter component performance, such as laser output as a function of drive current, laser MTTF, etc.BRIEF DESCRIPTION OF THE DRAWINGS

[0016] FIG. 1 is a schematic diagram of one embodiment of an artificial intelligence-enabled optical particle detector.

[0017] FIG. 2 is a schematic diagram showing an overview of a purified water system for use in a cleanroom, including the major unit operations thereof.

[0018] FIG. 3 is a schematic diagram showing the flow scheme for a purified water system in a manufacturing facility.STATEMENTS REGARDING CHEMICAL COMPOUNDS AND NOMENCLATURE

[0019] In general, the terms and phrases used herein have their art-recognized meaning, which can be found by reference to standard texts, journal references and contexts known to those skilled in the art. The following definitions are provided to clarify their specific use in the context of the invention.

[0020] “Particles” refers to small objects which are often regarded as contaminants. A particle can be any material created by the act of friction, forLeydig Ref. 340550: 29-25 WOexample when two surfaces come into mechanical contact and there is mechanical movement. Particles can be composed of aggregates of material, such as dust, dirt, smoke, ash, water, soot, metal, oxides, ceramics, minerals, or any combination of these or other materials or contaminants. “Particles” may also refer to biological particles, for example, viruses, spores and microorganisms including bacteria, fungi, archaea, protists, other single cell microorganisms. In some embodiments, for example, biological particles are characterized by a size dimension (e.g., effective diameter) ranging from 0.1-15 pm, optionally for some applications ranging from 0.5-5 pm. A particle may refer to a small object which absorbs, emits or scatters light and is thus detectable by an optical particle counter. As used herein, “particle” is intended to be exclusive of the individual atoms or molecules of a carrier fluid, for example water, air, process liquid chemicals, process gases, etc. In some embodiments, particles may be initially present on a surface, such as a tools surface in a microfabrication facility, liberated from the surface and subsequently analyzed in a fluid. Some systems and methods are capable of detecting particles comprising aggregates of material having a size dimension, such as effective diameter, greater than 20 nm, 30 nm, 50 nm, 100 nm, 500 nm, 1 pm or greater, or 10 pm or greater. Some embodiments of the present invention are capable of detecting particles having a size dimension, such as effective diameter, selected from that range of 1.0 nm to 150 pm, optionally for some applications 10 nm to - 10 pm, optionally for some applications 10 nm to - 1 pm, and optionally for some applications 10 nm to - 0.5 pm.

[0021] The expression “detecting a particle” broadly refers to sensing, identifying the presence of, counting and / or characterizing a particle, such as characterizing a particle with respect to a size dimension, such as effective diameter. In some embodiments, detecting a particle refers to counting particles. In some embodiments, detecting a particle refers to characterizing and / or measuring a physical characteristic of a particle, such as effective diameter, cross sectional dimension, shape, size, aerodynamic size, or any combination of these. In some embodiments, detecting a particle is carried out in a flowing fluid, such as gas having a volumetric flow rate selected over the range of 0.05 CFM to 10 CFM, optionally for some applications 0.1 CFM to 5 CFM and optionally for some applications 0.5 CFM to 2 CFM. In some embodiments, detecting a particle is carried out in a flowing fluid,Leydig Ref. 340550: 29-25 WOsuch as liquid having a volumetric flow rate selected over the range of 1 to 1000 mL / min. In some aspects, the systems and methods of the present disclosure may be employed in conjunction with non-optical particle detection techniques. In one aspect, an Al-enabled particle counter may comprise an acoustic wave particle detector. An acoustic wave particle detector may utilize acoustic induced cavitation and / or scattered acoustic energy signals for particle detection. Additionally, an acoustic wave particle detector may utilize, e.g., surface acoustic waves (SAWs) or Rayleigh waves, to detect and potentially manipulate particles. In another aspect, an Al-enabled particle counting method may comprise passing a particle-containing fluid though a nanopore media, thereby producing an ionic current as a sensor signal to be analyzed or filtered in order to detect particles.

[0022] “Optical Particle Counter,” “particle counter,” “particle detector” or optical particle analyzer are used interchangeably and refer to a particle detection system that uses optical detection to detect particles, typically by analyzing particles in a fluid flow. Optical particle counters include liquid particle counters and aerosol particle counters, for example, including systems capable of detecting individual single particles in a fluid flow. Optical particle counters provide a beam of electromagnetic radiation (e.g. a laser) into the analysis area, where the beam interacts with any particles and then detects the particles based on scatter, emitted or transmitted light from the flow cell. Detection may focus on electromagnetic radiation that is scattered, absorbed, obscured and / or emitted by the particle(s). Various detectors for optical particle counters are known in the art, including for example, single detection elements (e.g., photodiode, photomultiplier tube, etc.), detector arrays, cameras, various detector orientations, etc. Optical particle counter includes condensation particle counters, condensation nuclei counters, split beam differential systems and the like. When used in the context of a condensation particle counter, the particle counter portion refers to the detection system (e.g. source of electromagnetic radiation, optics, filters, optical collection, detector, processor, etc.). In an embodiment, for example, an optical particle counter comprises a source for generating a beam of electromagnetic radiation, beam steering and / or shaping optics for directing and focusing the beam into a region where a fluid sample is flowing, for example a liquid or gas flowing through a flow cell. A typical optical particle counterLeydig Ref. 340550: 29-25 WOcomprises of a photodetector, such as optical detector array in optical communication with said flow cell, and collection optics for collecting and imagining electromagnetic radiation which is scattered, transmitted by or emitted by particles which pass through the beam. Particle counters may further comprise electronics and / or processors components for readout, signal processing and analysis of electrical signals produced by the photodetector including current to voltage converters, pulse height analyzers, and signal filtering and amplification electronics. An optical particle counter may also comprise a fluid actuation systems, such as a pump, fan or blower, for generating a flow for transporting a fluid sample containing particles through the detection region of a flow cell, for example, for generating a flow characterized by a volumetric flow rate. Useful flow rates for samples comprising one or more gases include a flow rate selected over the range of 0.05 CFM to 10 CFM, optionally for some applications 0.1 CFM to 5 CFM and optionally for some applications 0.5 CFM to 2 CFM. Useful flow rates for samples comprising one or more liquids include a flow rate selected over the range of 1 to 1000 mL / min. The following references relate to optical particles counters and methods of making and using optical particles counters, which are hereby incorporated by reference to the extent not inconsistent with the present description, U.S. Patent nos. 4,893,928; 5,282,151; 6,859,277 ; 7,456,960; 7,746,469; 7,973,929; 8,154,724; 9,989,462; 10,371,620; 10,921,229; 11,181,455; 11,237,095; 11,320,360; 11,781,965;11,988,593 and 12,276,592 and U.S. Patent Publication Nos. US 2023 / 0236107; 2023-0087059; and 2024 / 0264066

[0023] Detecting and counting small particles (e.g., effective diameter less than 100 nm) in clean and ultraclean fluids in a manner that provides statistically significant data requires high signal-to-noise ratio (S / N). A high S / N ratio allows nanoparticles to be clearly detected above the noise floor. As used herein “statistically significant data” refers to detection of enough particles per unit time to be able to accurately assess contamination levels in the fluid. In some embodiments, high S / N does not relate to sizing accuracy directly. For example, in some optical particle counters the beam waist occupies a small fraction of the flow cell channel, and therefore, this approach monitors a subset of the total flow, such that it is possible for particles to pass through the edge of the beam where irradianceLeydig Ref. 340550: 29-25 WOis less than the center. If a 50 nm particle passes through the outer edge of the beam, it may generate a signal similar to a 10 nm particle passing through the center of the beam. Therefore, it is possible for some optical particle counters to have high S / N and be able to detect, for example 20 nm particles, while not having very good sizing accuracy. In some of the present optical particle counters and methods a goal is to be able to count enough particles to provide a quantitative, statistically sound, assessment of contamination levels in ultrahigh purity fluids in the shortest period of time. For example, the current state of the art particle counter may require up to 40 minutes to count enough particles to provide a statistically appropriate concentration (acceptable relative standard deviation) measurement when monitoring a state of the art ultrapure water system. By improving and maintaining a high S / N through the present systems and methods, the time interval needed to measure this minimum statistically acceptable particle counts can be reduced by 10x or more. This provides value as it allows a user to identify deviations from process control limits more quickly.

[0024] The term “noise” refers to unwanted modifications of a signal (e.g. a signal of a photodetector) that interfere with the accuracy or precision of a particle detection system. Noise may derive from sources such as backscatter due to interfaces between materials in the optical path of the beam, backscatter due to contamination of optical components, and / or molecular scatter from the fluid in the particle interrogation region from reaching the laser. Noise may also originate in electronic components or may be due to environmental instabilities. In some embodiments, noise due to backscatter may result in abnormal electronic signals whose amplitude can exceed a particle detection threshold, resulting in false-positive particle detection events.

[0025] The expression “high signal-to-noise ratio” refers to a signal-to-noise ratio of an optical particle detection system sufficient for accurate and sensitive detection of particles in a fluid flow, including particles characterized by a small physical dimension (e.g., an effective diameter of less than or equal to 200 nm, optionally for some embodiments less than or equal to 100 nm and optionally for some embodiments less than or equal to 50 nm). In an embodiment, “high signal-to-noise ratio” refers to a signal-to-noise ratio sufficiently high to sense particlesLeydig Ref. 340550: 29-25 WOcharacterized by a small physical dimension, such as particles having an effective diameter as low as 20 nm, optionally for some applications a diameter as low as 10 nm and optionally for some applications a diameter as low as 1 nm. In an embodiment, “high signal-to-noise ratio” refers to a signal-to-noise ratio sufficiently high to accurately detect and count particles with a false detection rate of less than or equal to 50 counts / L, for example, for detection of particles having an effective diameter selected over the range of 1 - 1000 nm. In an embodiment, “high signal-to-noise ratio” refers to a signal-to-noise ratio sufficiently high to provide a minimum statistically acceptable particle count in a timeframe at least a factor of 10x less than in a conventional optical particle counter. Systems and methods of the present disclosure may provide a high signal to noise ratio.

[0026] “Optical communication” refers to components which are arranged in a manner that allows light to transfer between the components.

[0027] "Light source" refers to a device or device component that is capable of delivering a beam of electromagnetic radiation to a sample. The term “light” is not limited to visible radiation, such as by a visible light beam, but is used in a broad sense to include any electromagnetic radiation also inclusive of visible radiation, ultraviolet radiation and / or infrared radiation. The light source may be embodied as a laser or laser array, such as a diode laser, diode laser array, diode laser pumped solid state laser, LED, LED array, gas phase laser, laser oscillator, solid state laser, to name a few examples.

[0028] The term “electromagnetic radiation” and “light” are used synonymously in the present description and refer to waves of electric and magnetic fields.Electromagnetic radiation useful for the methods of the present invention includes, but is not limited to ultraviolet light, visible light, infrared light, or any combination of these having wavelengths between about 100 nanometers to about 15 microns.

[0029] The term “detection region” refers to a zone within a particle detection system where one or more particles interact with the incident beam and / or the pump beam to scatter light. In some embodiments, the detection region may comprise a cuvette and / or a flow cell to constrain a particle-containing liquid flowing therethrough. In other embodiments, an unconstrained jet of particle-containing gasLeydig Ref. 340550: 29-25 WOmay flow through the detection region. In still other embodiments, the detection region may comprise a surface to be interrogated for particles.

[0030] The term “false-positive particle detection event” refers to an error in a particle detection system in which a signal is sent to a user, wherein the signal incorrectly indicates the presence of one or more particles in the particle detection region. In one aspect, a false-positive particle detection event can be caused by a noise event in the particle detection system. In one aspect, a false-positive particle detection event may include the particle detection system sending a signal to a user, that signal incorrectly indicating the presence of a single particle in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In one aspect, a false-positive particle detection event may include the particle detection system sending a signal to a user, that signal incorrectly indicating the presence of a plurality of particles in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In one aspect, a false-positive particle detection event may include the particle detection system sending a signal to a user, that signal incorrectly indicating the presence of an unusually high concentration of particles in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In other aspects, a false-positive particle detection event may include the particle detection system sending a signal to a user, that signal incorrectly indicating a higher concentration of particles than is accurate. In one aspect, a false-positive particle detection event may be caused by one or more signals generated by a photodetector of the particle detection system, wherein the one or more signals exceed a threshold value, the threshold value being preselected to be sufficiently high to distinguish light associated with the presence of a particle from ordinary background noise received by and / or generated in the photodetector.

[0031] The term “noise event” refers to an error in a particle detection system in which a signal is generated by a photodetector of a particle detection system and sent to a processor of the particle detection system, wherein the signal incorrectly indicates the presence of one or more particles in the particle detection region. In one aspect, the signal generated by the photodetector and sent to the processorLeydig Ref. 340550: 29-25 WOmay have a plurality of channels, each channel corresponding to a particle size bin. A noise event may occur on just one channel or several channels of the signal generated by the photodetector. In one aspect, a noise event may include the photodetector generating a signal and sending the signal to the processor, that signal incorrectly indicating the presence of a single particle in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In one aspect, a noise event may include the photodetector generating a signal and sending the signal to the processor, that signal incorrectly indicating the presence of a plurality of particles in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In one aspect, a noise event may include the photodetector generating a signal and sending the signal to the processor, that signal incorrectly indicating the presence of an unusually high concentration of particles in the particle detection region, when in fact there were no particles in the particle detection region at the point in time associated with the signal. In one aspect, a noise event may be caused by a signal being erroneously generated by a photodetector of the particle detection system, wherein the one or more signals exceed a threshold value, the threshold value being preselected to be sufficiently high to distinguish collected light associated with the presence of a particle from ordinary background noise received by and / or generated in the photodetector. Thus, an erroneous false-positive particle detection signal may be generated by the photodetector an sent to the processor in a noise event. In one aspect, a noise event may be caused by a misalignment between the photodetector and a scattered light focusing system. In one aspect, a noise event may be caused by focused scattered light from bulk liquid being transmitted to the photodetector. In one aspect, a noise event may be caused by scattered light that extends beyond the active photodetector area causing reflection in the detector. In one aspect, a noise event may be caused by a laser RMS noise level being too high. In one aspect, a noise event may be caused by an abrupt change in laser output. In one aspect, a noise event may be caused by an oscillation or change in the beam propagation axis of the laser. In one aspect, a noise event may be caused by an abrupt change in DC light level and / or background noise.Leydig Ref. 340550: 29-25 WO

[0032] The term “background noise” means noise of the particle detection system that produces a signal that falls below a predetermined threshold value, the threshold value being preselected to be sufficiently high to distinguish collected light associated with the presence of a particle from the ordinary background noise.Thus, in one aspect, background noise may be distinguished from a noise event in that the background noise does not cause a false-positive particle detection event, whereas a noise event does.

[0033] The term “noise signature” means one or more characteristics of the signal of a noise event that may be used to distinguish the noise event from a signal caused by the presence of a particle in the particle detection region. In one aspect, a noise signature may comprise signal characteristics that include one or more ratios of the signal of one channel to that of another channel. For example, it has been observed that some noise events involve the generation of a signal on the first channel incorrectly indicating the presence of a high concentration of particles of the smallest particle size bin, accompanied by signals of other channels for larger particle size bins falling well below the concentration of particles that would be expected for the larger sizes given the high signal of the first channel. The ratios may be calculated over one time interval or many time intervals. In one aspect, a noise signature may comprise signal characteristics wherein the raw particle count for a short sequence of elemental data intervals (e.g., 1-5 elemental data intervals of 100ms each) is large with respect to neighboring elemental data intervals in a sequential series of elemental data intervals produced by the photodetector. This is based on the observation that correct particle counts generally do not instantaneously or nearly instantaneously jump to high levels, potentially due to physical phenomena like particle dispersion in the bulk liquid. In one aspect, a noise signature may comprise signal characteristics wherein for a relatively longer sequence of elemental data intervals, the total raw particle count is too large with respect to a baseline average and / or historical data. The noise signature may comprise signal characteristics which can be discerned via comparison of short, medium, and long sequences of elemental intervals. In one aspect, a noise signature may comprise signal characteristics wherein absolute rates of change of a raw particle count trend line are too high. In one aspect, a noise signature mayLeydig Ref. 340550: 29-25 WOcomprise signal characteristics wherein comparison of raw particle counts across several channels (e.g., channels corresponding to 20, 50, 70, and 100nm counts) does not meet specific distribution thresholds. In one aspect, a noise signature may comprise signal characteristics wherein a finite data interval (10 sec to 10 min) having high particle counts is bounded on one or both sides by normal data intervals with normal counts (i.e. roughly equal to the baseline) where all counts in the finite data interval are greater than the baseline.

[0034] The term Taw particle count data” refers to unaltered data embodied in the signal produced by the photodetector of a particle detection system. As such, raw particle count data will include any erroneous particle counts resulting from one or more noise events.

[0035] The term “particle measurement output” means particle count information sent from the particle detection system to a user. In some embodiments, the particle count measurement may be sent at a reporting interval, the particle count information being based on the filtered particle count data. In one aspect, the particle count information may include particle counts associated with different particle size bins, as derived from the different channels of the photodetector signal. Thus, the particle count information may include information relating to particle concentration and or size distribution over the reporting interval. The particle measurement output may also include a filtering report that indicates at least one of the following: whether any data was removed from the raw particle count data as a result of filtering over the time period associated with particle measurement, whether any replacement data was used to fill gaps in the data, the total time durations of the replacement data used, and / or how the replacement data was derived.

[0036] The term “sensor signal” refers to the electrical output generated by a detector, for example a photodetector, in response to particles passing through a beam of electromagnetic radiation. A sensor signal may represent the intensity of light scattered or absorbed by particles which pass through the beam of electromagnetic radiation as measured in volts. The sensor signal may vary in terms of amplitude or duration based on the characteristics of the particles detected by the beam of electromagnetic radiation. For example, the sensor signal may vary basedLeydig Ref. 340550: 29-25 WOon the size, shape, or concentration of particles. The sensor signal may be used to quantify or otherwise analyze the particles present in a sample. Interpretation of the sensor signal enables accurate measurement and classification of particles for various applications. In some embodiments, the sensor signal may originate from a sensor of the detector, e.g., a photodetector of the detector, as an analog electrical signal. Thus, in some embodiments, processing the sensor signal comprises converting the analog sensor signal to a digital signal in order to generate the artificial intelligence-enabled particle counter output.

[0037] The term “artificial intelligence model” refers to a computational framework or algorithm enabling computer understanding, using machine learning, deep learning, or artificial neural networks, to replicate a cognitive decision process. An artificial intelligence model may be trained on datasets to recognize patterns, make predictions, or derive insights. In the context of optical particle counters, an artificial intelligence model may be employed to enhance performance by facilitating tasks such as data analysis, anomaly detection, classification of particles, and adaptive calibration.

[0038] The term “artificial intelligence-enabled particle counter output” refers to new data or data structures that an artificial intelligence model creates or synthesizes based on input data and the artificial intelligence model’s algorithm. The output data can take various forms including, but not limited to, numerical values, categorical classifications, data transformations, alarms, or actionable recommendations depending on the nature of the task and the design of the artificial intelligence system.

[0039] The term “particle measurement output” refers to the quantitative and qualitative data generated by a particle measurement system, such as an optical particle counter, following the detection and analysis of particles in a sample. Particle measurement output can include parameters such as particle count, particle size distribution, concentration, composition, biological viability, or particle morphology. Particle measurement output may be derived from the processing of raw signal data obtained from the detection process. Particle measurement output providesLeydig Ref. 340550: 29-25 WOinformation regarding the presence and characteristics of particles in the analyzed sample.

[0040] The term “system data” refers to information generated and collected by an optical particle counter and / or one or more sensors external to the particle counter that characterizes the operational state, performance characteristics, user interface settings, ambient environmental conditions, configuration of the optical particle counter. System data may also refer to the comprehensive set of information input to an optical particle counter. Examples of system data may include internal parameters such as raw particle count measurements, particle size distribution data, environmental conditions such as temperature and humidity, calibration parameters, operational parameters such as flow rates and sampling times, laser parameters, including but not limited to laser power, laser stability, and laser condition and optical bench parameters. Internal system data may also include photodetector parameters, including but not limited to photodetector signal, photodetector biasing, photodetector noise, and photodetector background. System data may also include external parameters such as user input, historical performance records, maintenance records, and any additional related information that can be utilized to train artificial intelligence algorithms. The use of such system data in training artificial intelligence algorithms may allow for, for example, improved pattern recognition, predictive analysis, and improvements in the accuracy, sensitivity, and overall reliability while also facilitating, for example, real-time diagnostics, performance optimization, and quality control of optical particle counters.

[0041] The term “probability map” refers to an output that gives the probability that the variable of interest is above or below a specified threshold value. Thus, in some embodiments, a probability map may be a multidimensional (e.g., 2, 3, 4+ dimensions) surface that represents the probability that the variable of interest is above or below the threshold value. In one particular embodiment, a probability map may be a three-dimensional surface plotted in a three dimensional space, the three dimensions comprising: (1 ) time and (2) particle size bins (or channels) of a particle detector, as independent variables, and (3) the probability that a particle was correctly detected (as opposed to, e.g., a noise causing a false detection) as theLeydig Ref. 340550: 29-25 WOthird dimension / variable of interest. A probability map may or may not include location as an independent variable / axis.

[0042] The term “cloud-based” refers to Al models and associated databases that utilize remote servers for storing data and applications, which may be accessed over the internet. Thus, users of could-based Al applications may not need to install at least a portion of the software locally, instead accessing it through a web browser or application.

[0043] The expression “interferometric detection of particles” refers to systems and methods using optical interference to detect one or more particles. In some embodiments, coherent beams of electromagnetic radiation are superimposed to cause optical interference for sensing, counting and / or determining a size characterization of a particle that interacts with at least a portion of the electromagnetic radiation.

[0044] The expression “differential detection” refers to techniques and systems using the differential signal from forward looking on axis detector pair(s) for example, at a scattering angle that is within 0.5 degrees of zero degrees relative to the optical axis of the incident beam, optionally for some applications at a scattering angle that is within 0.1 degree of zero degrees relative to the optical axis of the incident beam, and optional is at or near zero degrees. A minimum of two pixels can be used to generate a differential signal (e.g., one upper (or top) and one lower (or bottom)), thereby forming a single pixel pair for differential detection. Alternatively, a plurality of pixels may be employed for each active detector area of a differential detector (e.g., the top active region and the bottom active region), such as a segmented differential detector comprising one or more pixel pairs, thereby using a plurality of pixel pairs, for example, wherein one pixel of each pixel pair corresponds to top active detector region and the other pixel of each pixel pair corresponds to the bottom active region. The number of pixel pairs may range, for example, from 1 to 500 pixels and, and optionally for some applications from 50-100 pixels. In some embodiment, the differential signal is generated by differentially adding signals from pixel pairs corresponding to different active regions of a segmented detector array, such as the top half and the bottom half. Differential detection may be used in the presentLeydig Ref. 340550: 29-25 WOsystems and methods to provide a reduction of noise and thus enhanced signal-to-noise ratio. In some embodiments, for example, differential detection is used for detection of the combination of incident electromagnetic radiation transmitted through said flow cell and electromagnetic radiation forward scattered by one or more particle(s) in the fluid flow in the flow cell. In some embodiments, for example, the distribution of light incident has a power distribution that is balanced between the first and second active detection regions (e.g., upper half and lower half) of the differential detector, for example, such that the first and second active detection regions are characterized by incident radiant powers that are within 10%, optionally for some applications 5% and optionally for some application within 1%. Differential detection includes techniques and systems having closed loop control, for example, based on an evaluation the noise amplitude of the differential signal when particles are not present (i.e. , in the absence of scattering form the particle). In some embodiments, a steering mirror is used to adjust the incident beam position on the detector to reduce or minimize noise levels of the differential signal, which may occur when the beam power is uniformly split between the first and second active detector elements (e.g., upper and lower elements of the detector). Close loop control can also be achieved by translating the detector position and rotating the detector to align the beam and detector axes so as to reduce or minimize noise levels of the differential signal.

[0045] “Structured Beam” refers to a coherent beam of electromagnetic radiation (e.g., a laser) having a non-Gaussian spatial intensity distribution. Structured beams include beams characterized by an attenuated region, such as a dark beam, beams with a line focus with a dark line singularity, beams characterized by two or more discrete intensity lobes, etc. In an embodiment, a structured beam corresponds to a transverse mode, such as a TEM01. Structured beams include focused, synthesized, laser beams. Structured beams and dark beams may be generated by techniques known in the art including using an optical mask, modification of a laser cavity, combining multiple beams, spatial and / or polarization filters and other manipulations such as in an interferometric or polarization modification scheme.

[0046] “Beam propagation axis” refers to an axis parallel to the direction of travel of a beam of electromagnetic radiation.Leydig Ref. 340550: 29-25 WODETAILED DESCRIPTION OF THE INVENTION

[0047] In the following description, numerous specific details of the devices, device components and methods of the present invention are set forth in order to provide a thorough explanation of the precise nature of the invention. It will be apparent, however, to those of skill in the art that the invention can be practiced without these specific details.Turning now to FIG. 1, a method for artificial intelligence-enabled particle detection may comprise passing a particle-containing fluid 150 through a detection region 210 of an optical particle detector. The fluid in the detection region may then be exposed to a beam of electromagnetic radiation 221 produced by light source 220, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles being collected by collection optic 230. At least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles may be directed, via the collection optic 230 onto a photodetector 240; generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector. The sensor signal, or a data set based thereon, may then be processed in controller 101 using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output. A particle measurement output may then be determined, wherein the particle measurement output is based at least in part on the artificial intelligence-enabled particle counter output.

[0048] The invention can be further understood by the following non-limiting examples.

[0049] Example 1: Unit Matching

[0050] To attain reliable data interpretation between multiple particle counters in multiple locations, particle counter operational algorithms are trained using the same machine learning model and same fluid. There are factors which can causeLeydig Ref. 340550: 29-25 WOvariability between particle counters when measuring particles in the same fluid. These factors can include, for example, beam shape, beam size, flow rate, environmental variability, flow rate variability, differences in calibration, drift in calibration accuracy overtime, noise levels induced by specific components, etc.

[0051] Optimized matching of particle counters occurs when both the counting of real particles passing through the detection region and the rejection of noise / false counts are consistent across multiple particle counter units. A classification model is used to consistently distinguish between real and false counts, using supervised, unsupervised, or semi-supervised learning. Regression models are used to tune detection and sample volume parameters to obtain similar performance across multiple particle counters.

[0052] As a result, particle counter measurements are optimally matched such that particle measurements performed in different locations, different facilities, and even on different continents can be compared with confidence. The learning models are optimized for individual customers or specific chemicals or applications, such that additional particle counters added later share the same model and same benefits.

[0053] Example 2: Machine Learning for Systemic Noise Reduction in Particle Counting

[0054] When monitoring ultra-clean fluids, such as those used in semiconductor or pharmaceutical manufacturing, particle detection rates can be extremely low. In these conditions, even a small number of false positives — resulting from, for example, electronic noise, laser scattering anomalies, microbubble interference, or mechanical misalignment — can significantly distort the measurement and lead to inaccurate assessments of fluid cleanliness.

[0055] Machine learning (ML) offers a powerful toolset to improve measurement accuracy by intelligently differentiating true contamination events from systemic or environmental noise. By continuously analyzing high-dimensional sensor data streams, ML algorithms learn to recognize characteristic signal patterns of genuine particles versus those stemming from artifacts such as:Leydig Ref. 340550: 29-25 WO

[0056] Electronic interference: transient spikes or drifts in signal baseline due to thermal or component-level fluctuations;

[0057] Laser noise: beam instability, mode hopping, or scattered light interactions with internal surfaces;

[0058] Microbubble introduction: sporadic, high-scattering events with differing shape or duration profiles compared to hard particles;

[0059] Opto-mechanical misalignments or vibration-induced jitter: which can modulate the optical path and induce spurious scattering events.

[0060] Using historical and real-time data, ML models — such as supervised classifiers (e.g., random forests, SVMs) or unsupervised techniques (e.g., anomaly detection algorithms, autoencoders) — are trained to:

[0061] Identify and suppress known noise signatures;

[0062] Adapt to changing baseline conditions over time;

[0063] Enhance signal-to-noise ratio by weighting or denoising inputs;

[0064] Flag unusual patterns that may indicate system drift, requiring recalibration or maintenance.

[0065] Crucially, these models are embedded in edge-processing architectures, allowing for low-latency classification of events, minimizing false positives in near-real-time. By reducing system noise and rejecting false counts, ML-enhanced particle counters are able to detect smaller particles with greater sensitivity and reliability — particularly valuable in applications where regulatory compliance or yield-critical decisions depend on precise contamination data.

[0066] In summary, integrating machine learning into particle counting systems offers a transformative opportunity to enhance performance through intelligent noise suppression and / or adaptive signal discrimination.

[0067] Example 3: Enhancing Particle Counter Sensitivity with Machine Learning- Based Pattern RecognitionLeydig Ref. 340550: 29-25 WO

[0068] In traditional optical particle counters, the detectability of the smallest particles is limited by a fixed voltage threshold. This threshold is typically set several standard deviations above the system’s baseline noise to avoid false positives. While effective for minimizing erroneous counts, this approach inherently excludes low-amplitude particle events that fall near or below the set threshold — particularly problematic in ultra-clean environments where contamination levels are low and detecting sub-1 OOnm particles is critical.

[0069] Machine learning (ML) introduces a paradigm shift by replacing static thresholds with dynamic, pattern-based detection models. Instead of relying solely on signal amplitude, ML algorithms can analyze the full temporal and spectral signature of detected events, including:

[0070] Pulse shape and duration

[0071] Rise and fall times

[0072] Signal-to-noise ratio

[0073] Frequency-domain characteristics

[0074] Cross-channel correlations (e.g., from multiple photodetectors)

[0075] By training supervised models (e.g., convolutional neural networks, gradient boosting machines) or leveraging unsupervised techniques (e.g., clustering, autoencoders), the system learns the statistical fingerprint of true particle signals versus various types of noise — such as shot noise, microbubble scattering, vibration-induced misalignment signals, or electrical transients. This allows for the accurate classification of events that would otherwise be discarded using fixed thresholds.

[0076] More specifically, ML enables:

[0077] Adaptive thresholding: ML models continuously adjust detection criteria based on real-time system behavior and historical patterns, allowing for lower thresholds without increasing false positives.Leydig Ref. 340550: 29-25 WO

[0078] Event pattern recognition: Instead of reacting to amplitude alone, ML models consider multidimensional signal features to identify the complex temporal patterns typical of genuine particle interactions.

[0079] Noise signature suppression: By learning the unique characteristics of different noise sources (e.g., thermal noise vs. optical scatter), models isolate and subtract noise components from raw signal data, enhancing the purity of the detected particle signal.

[0080] False count minimization: ML flags and discards anomalous signals that traditional threshold-based systems might misclassify, such as reflections, air bubbles, or photodiode dark noise fluctuations.

[0081] Improved sensitivity at smaller sizes: As noise rejection becomes more sophisticated, detection can operate closer to the physical noise floor of the system, enabling the identification of smaller particles that traditional systems would miss.

[0082] In essence, ML-based signal analysis transforms the particle counter from a threshold-dependent detector into a pattern recognition engine. This not only increases the confidence in detected events but also meaningfully extends the lower limit of detection — an essential advantage in applications demanding extreme cleanliness, such as in clean room applications for advanced semiconductor manufacturing or sterile pharmaceutical production.

[0083] Moreover, these models are retrained or fine-tuned over time, allowing the system to adapt to component aging, fluid chemistry shifts, or environmental changes without requiring manual recalibration.

[0084] Example 4: Biological Identification - Application of Machine Learning Models to Intrinsic Fluorescence Particle Counting for Viable Microbial Discrimination

[0085] In many high-purity applications, such as aseptic pharmaceutical manufacturing or sterile filling operations, it is not sufficient to merely quantify the number of particles in a fluid or air stream. It is equally critical to identify the nature of those particles — particularly differentiating viable microbial contaminants from non-biological particles or non-viable biological debris. Among the available techniques,Leydig Ref. 340550: 29-25 WOintrinsic fluorescence has emerged as a powerful tool for detecting viable organisms due to their natural emission response to UV illumination.

[0086] However, this method is prone to false positives from non-biological materials that exhibit similar fluorescent behavior — such as optical brightening agents, detergent residues, or certain polymer fragments. These materials may fluoresce under UV light but do not indicate microbial presence or viability. To address this, machine learning (ML) is used to dramatically improve the selectivity and reliability of detection systems through advanced signal classification.Multi-Detector & Multi-Spectral Signal Capture

[0087] Modem particle counters designed for biological discrimination often incorporate multiple photodetectors, configured to capture:

[0088] Elastic scatter (e.g., 90°, forward, and back scatter)

[0089] Fluorescence emission across multiple wavelength bands (e.g., 420-460 nm, 500-550 nm, etc.)

[0090] Fluorescence intensity over time (rise, dwell, and decay characteristics)

[0091] These instruments generate high-dimensional datasets for each particle event, including:

[0092] Relative emission intensities at different wavelengths

[0093] Temporal signal characteristics such as rise time, peak amplitude, duration, and decay slope

[0094] Signal-to-noise ratio and baseline shift

[0095] Scatter-emission correlation patternsRole of Machine Learning in Signal Discrimination

[0096] Machine learning is applied across several layers of this detection process:Leydig Ref. 340550: 29-25 WO

[0097] Spectral Feature Analysis: As disclosed herein, ML models differentiate viable microbial signatures from false positives by analyzing spectral emission ratios. For example, viable bacteria may produce a characteristic emission peak at 445-465 nm due to metabolic fluorophores (e.g., NADH), whereas non-biological fluorophores often have broader or shifted emission spectra.

[0098] Temporal Signature Differentiation: Neural networks or time-series classifiers are trained to recognize subtle differences in fluorescence kinetics. Viable organisms often produce fluorescence with unique rise and decay profiles, influenced by intracellular fluorophore interactions, while brightening agents typically have sharper, more symmetric peaks.

[0099] Multi-Channel Correlation: As disclosed herein, ML simultaneously analyzes signals from multiple detectors — comparing scatter and fluorescence profiles to evaluate if the event corresponds to a biological particle (which often exhibits both) or a purely fluorescent artifact (which may lack correlated scatter).

[0100] Viability Classification: Using supervised learning (e.g., ensemble decision trees or support vector machines), the system is trained on labeled data to classify incoming events as:

[0101] Viable microbial particles

[0102] Non-viable biological debris

[0103] Non-biological fluorescent particles

[0104] Background noise or electronic artifacts

[0105] Data Fusion and Predictive Scoring: As disclosed herein, Bayesian inference or deep learning models aggregate evidence across multiple signal dimensions to produce a probability score of microbial viability per event — rather than relying on binary threshold logic.Advantages of Machine Learning-Enhanced DiscriminationLeydig Ref. 340550: 29-25 WO

[0106] Higher specificity: ML reduces false positive rates caused by materials like whitening agents, ensuring that only biologically relevant events are flagged for action.

[0107] Increased sensitivity: By learning from true microbial emission profiles, the system detects low-emission or metabolically weak organisms that might otherwise be missed.

[0108] Adaptive behavior: Models are periodically retrained with updated contamination data, allowing for evolving understanding of new materials or environmental backgrounds.

[0109] Speciation potential: Overtime, with sufficient labeled data, ML distinguishes between different microorganism types based on their fluorescenceemission profiles and scatter signatures (e.g., bacterial vs. fungal).

[0110] Example 5: Intra-Device Data Fusion

[0111] Typical particle measuring systems utilize a single narrow wavelength laser to illuminate the sample volume under test. The scattered light from the sample volume may be measured by several different sensors to increase the dynamic range of particle (contamination) sizes that can be discriminated.Additional sensors types and lasers with different wavelengths are added to particle detection systems to provide an expanded data set of information that feeds Al algorithms improving their ability to:

[0112] Separate particle (contamination) generated signals from noise sources including scattered light generated from the bulk material under test.

[0113] Compensate for changes in environmental or internal component operating conditions.

[0114] Identify particle characteristics by how the particle is affected by the measurement environment and how the particle affects the environment.

[0115] Adapt to different materials under test.Leydig Ref. 340550: 29-25 WO

[0116] Identify when and by how much baseline operations have changed.

[0117] Measure and identify changes in the bulk material under test.

[0118] Sensors are used to measure the material under test from different angles or measurement conditions, changes in environmental conditions, or changes in operating conditions.

[0119] Additional sensor types, as disclosed herein, include:

[0120] Temperature & Pressure (Compensate for changes in bulk material viscosity)

[0121] Power (AC & DC) (Compensate for changes in operating conditions)

[0122] Flow (Compensate for turbulence and changes in the dwell time through the measurement volume or differences in velocity with respect to the bulk material under test)

[0123] Electric Field Sensing (Detect the presence of metal contaminants).

[0124] Chemical (Detect changes in chemical properties of the material under test). For example, known chemicals are injected into test flow to react with contaminates.

[0125] Light sensors to measure changes in laser power (Compensate for changes in output intensity (instantaneous, mode hopping, or gradual trending)

[0126] Video cameras (High speed mono or multispectral) (Measure contaminate flow through sample volume) Video cameras also provide size, trajectory, and shape information.

[0127] Using Laser with different wavelengths either simultaneously illuminating or sequentially illuminating a sample volume from the same or different attack angles.

[0128] Example 6: Transforming User Experience with Machine Learning-Driven Adaptive User InterfacesLeydig Ref. 340550: 29-25 WO

[0129] In advanced optical particle counters and their associated software platforms, the number of available features and configuration options are extensive. While this breadth of functionality supports flexibility and application diversity, it can also lead to user friction — especially when navigating complex menus, searching for frequently used settings, or performing routine workflows undertime pressure.

[0130] Machine learning (ML) offers a transformative approach to optimizing the human-machine interface by enabling adaptive user interfaces (AU Is) that dynamically tailor themselves to each user’s unique behavior patterns, preferences, and workflows. This creates a more intuitive, efficient, and personalized experience.

[0131] Key Components of a Machine Learning-Enabled Adaptive Ul

[0132] Feature Usage Tracking

[0133] The system continuously logs interactions, such as: frequently accessed functions, typical configuration settings, navigation paths taken through the Ul, time spent on specific screens.

[0134] Usage data can be linked to a user ID or role, ensuring preferences are retained across sessions and devices.

[0135] Behavioral Pattern Learning

[0136] Using unsupervised learning (e.g., clustering, dimensionality reduction), the system identifies common usage personas, such as: “Calibration specialists” who prioritize sensor diagnostics and alignment, “QA users” who focus on audit trails and data export, “Technicians” who frequently start / stop measurements and monitor live data.

[0137] For individual users, reinforcement learning or collaborative filtering models learn and predict preferred sequences of actions, time-of-day behaviors, or settings configurations.

[0138] Personalized Interface ReconfigurationLeydig Ref. 340550: 29-25 WO

[0139] Based on learned patterns, the III reorganizes itself to: promote most-used features to the top menu or home screen, pre-fill default values based on historical inputs, suggest likely next actions based on task history (predictive workflows), hide or collapse rarely used elements, reducing visual clutter.

[0140] Interfaces may be role-based (engineer, technician, manager) or even task-specific (calibration, validation, troubleshooting).

[0141] Context-Aware Hints and Automations

[0142] ML models are trained to trigger in-app guidance or automated workflows when they detect: users struggling with a multi-step task (e.g., prolonged dwell time or backtracking), deviation from normal process (e.g., missing a routine pre-check), opportunity to streamline (e.g., offering batch export when detecting multiple manual exports).

[0143] Cross-User Learning (With Privacy Controls)

[0144] Anonymized usage data across the user base can be aggregated to inform interface design improvements and feature prioritization.

[0145] New users can benefit from pre-configured templates derived from users with similar roles or workflows, accelerating onboarding.

[0146] Strategic Advantages

[0147] Increased Operational Efficiency: Users complete tasks faster with fewer clicks and less searching, particularly valuable in high-throughput or compliance-sensitive environments.

[0148] Improved Customer Satisfaction: Interfaces feel personalized and responsive, which reduces frustration and boosts perception of product quality and innovation.

[0149] Reduced Training Requirements: Adaptive Uls shorten learning curves and minimize the need for detailed user manuals or external support.Leydig Ref. 340550: 29-25 WO

[0150] Higher Feature Adoption: As users are nudged toward underutilized features that align with their needs, product value perception increases.

[0151] Product Differentiation: A behavior-aware interface adds tangible competitive advantage in markets with otherwise similar hardware specs.

[0152] Applications

[0153] Assuming a user who consistently applies specific configurations settings and accesses only a small portion of data displays. Over time, the system:

[0154] Pre-loads those configuration options

[0155] Reconfigures the display making common data readily available or providing access more easily than would otherwise be available through hierarchical menu selections

[0156] Provides single-click access to validation workflows from the home screen.

[0157] In this way, the instrument doesn't just support the user — it learns with them, anticipating needs, simplifying tasks, and delivering a seamlessly adaptive experience.

[0158] Additionally, information about specific measurement types is optionally provided to the user and can include baseline statistical information about contamination levels based on particle size or other characteristics, frequency and nature of previous alerts / alarms associated with this measurement, long term trending analytics, and non-contamination related information.

[0159] This capability may include, for example, hiding features not routinely used by a particular user, group or organization, highlighting features for ease of identification in the event that a trigger has occurred requiring user interaction, snap-to behavior to accelerate user workflows by reducing latency. For supervisors and quality departments this includes, for example, analytics on user performance and workflow improvements through automated SOP drafting.Leydig Ref. 340550: 29-25 WO

[0160] Generative Al, and LLMs, as disclosed herein, are used to adapt the user interface and user guide of complex instrumentation to accelerate user training. In particular the combination of use data from a specific organization and the instrument user guide and III is combined to develop organization and workflow specific training, optimizing the training efficacy as well as increasing the value of the training provided.

[0161] Example 7: Hardware Diagnostics / Predictive Analytics for Maintenance and Calibration

[0162] Instrument performance degradation over time may affect measurement reliability. As disclosed herein, machine learning is applied to analyze historical data trends enabling it to predict when instruments need maintenance or recalibration, reducing downtime and ensuring consistent sensitivity. This data may include diagnostic data related to lasers, photo detectors, flow sensors, environmental effects, etc.

[0163] As disclosed herein, Al is used to determine optimal maintenance, calibration, or warranty durations. Sensors are used to monitor the operating conditions of the unit to help evaluate how environmental conditions may be accelerating the degradation or changing the performance of system or system components. Al is used to provide analysis and recommendations on how to improve the system designs for better reliability and extended MTTF and MTTB durations.

[0164] Another source of rich data for model training and optimized calibration and maintenance procedures includes, but is not limited to, the service records and customer support artifacts. By mining, for example, the emails, test results, service invoices and other artifacts, trends and opportunities for improvement are identified, as well as informing spare parts usage. This approach facilitates improvements in time to resolution and reducing downtime.

[0165] The use of generative Al and talk to text is adopted in such models to improve service outcomes as well as enriching the data sources for future model generation without placing undue burden on service and customer support staff.Leydig Ref. 340550: 29-25 WO

[0166] This approach is also benefited from training on R&D data from the product development process, improving the knowledge transfer from R&D to production and service teams, often a bottleneck in ramp-up of a product to achieving customer value. This data is, for example, in the form of test results which can inform normal operating parameters and bounds, development notes which can inform work instructions and FAQs, which can accelerate troubleshooting.

[0167] Example 8: Al compensation for Hardware Cost Optimization

[0168] Typically, use of lower quality components results in poorer instrument performance and potentially, non-functional instrumentation. However, high quality measurement data is obtained from instrumentation employing low-cost components and simplified operations when machine learning models compensate for low cost / low performing instrument components. Manufacturing costs are reduced by using lower cost / lower quality components such as lasers, detectors, cameras, electronics, etc.

[0169] A machine learning model can learn the noise characteristics of a low cost / low quality laser, for example, and compensate for the presence of that noise such that it does not affect measurement quality, therefore, cost savings are realized without affecting instrument performance metrics. This provides a competitive advantage for the instrumentation manufacturer.

[0170] This approach may require input characterization of the components to inform the model creation and running but may also rely on generic parameters. Over time, this approach allows tolerances to be widened while maintaining performance, or tolerances to remain the same while improving performance.

[0171] Example 9: Adaptive Measurement Algorithms

[0172] Dynamic environmental conditions can affect contamination monitoring. As disclosed herein, machine learning models adapt measurement parameters on-the-fly, based on environmental conditions such as humidity and temperature, optimizing sensitivity and reliability in diverse scenarios. Other measurement related parameters, include, for example, ambient pressure, sample pressure, sample flow rate, etc.Leydig Ref. 340550: 29-25 WO

[0173] Instrument parameters such as detection threshold voltages, detector gain, implementation of specific algorithm types, and laser power are optimized dynamically to provide the best quality measurement data for an evolving set of measurement conditions.

[0174] As noise sources can originate from a wide range of environmental factors, adding sensors to the product to monitor environmental and non-particle attributes of the sample media provides means to detect and avoid false counts. Using simple sensors such as pressure, temperature, humidity, refractive index, dissolved gas, pH, conductivity, etc. in combination with the optical particle counting sensing by means of machine learning augments the sensitivity to true contaminant particles. For example, as disclosed herein, this is achieved through adaptively setting instrument parameters, data quality metrics informing user on interpretation, or other techniques.

[0175] Example 10: General Implementation Examples - Sensitivity Improvements via S / N Ratio Improvement

[0176] Example 10A:

[0177] A machine learning program was developed to detect signals from high-energy particles (such as cosmic rays) and distinguish them from contamination particle signals. The model was trained using two datasets: one containing only high-energy particle signals, and another composed primarily of regular particle signals with a few high-energy events. This training strategy enabled the model to learn the unique characteristics of high-energy particles as well as their occurrence in more realistic, mixed scenarios. By analyzing features such as amplitude, duration, and waveform patterns, the model accurately identifies high-energy events even in the presence of noise. This enhances detection reliability and minimizes false positives, providing an effective solution for particle analysis in experimental environments.

[0178] The use of convolutional methods in training proved to be highly effective due to their ability to capture local patterns in waveform data. In this application, high-energy particle signals often exhibit distinct localized features — such as sharper peaks or higher amplitudes over short durations — that are easily lost using traditionalLeydig Ref. 340550: 29-25 WOfully connected networks. Convolutional neural networks (CNNs) apply filters across the time-series input to detect such features regardless of their exact position in the signal. This spatial invariance and local feature sensitivity make CNNs particularly well-suited for identifying high-energy events embedded in noisy or variable-length pulse data. Furthermore, the hierarchical structure of CNNs allows early layers to detect basic signal shapes, while deeper layers capture more complex patterns, leading to more accurate and reliable classification of particle types.

[0179] Example 10B:

[0180] A new approach was explored using a Langevin-based model to denoise incoming particle scatter signals. The Langevin equation, commonly used to describe systems influenced by both deterministic forces and random fluctuations, offers a physics-inspired framework well-suited for modeling noise in particle detection. By treating the signal as a stochastic process, the model distinguishes the underlying physical response from random fluctuations introduced by molecular scatter or electronic noise. This provides a principled method for suppressing noise while preserving the essential signal features, making it a promising technique for improving the clarity and reliability of weak particle detection in high-sensitivity measurements.

[0181] In this context, the Langevin-based method functions as an unsupervised denoising approach. Rather than relying on labeled examples of clean and noisy signals, it leverages the stochastic formulation of the Langevin equation to model the temporal evolution of noise and its interaction with the signal. Assuming the observed data consists of a deterministic component superimposed with stochastic noise, the model iteratively refines the signal estimate using simulated Langevin dynamics. This enables effective noise suppression without prior knowledge of the clean signal, making it especially advantageous in situations where labeled data is unavailable. The unsupervised nature of the method enhances its adaptability for real-time and experimental environments.

[0182] Example 10C:Leydig Ref. 340550: 29-25 WO

[0183] Detecting weak particle scattering signals in noisy environments is a critical challenge in optical detection systems. Improving the signal-to-noise ratio (SNR) before digital processing is essential for reliable real-time detection. As disclosed herein, a combined analog and stochastic approach offers an effective solution.

[0184] A hybrid method is proposed that integrates phase modulation, lock-in amplification, and a Langevin-based de-noising model to enhance scattered particle signal detection. Phase modulation is applied to the laser pulse, enabling the lock-in amplifier to extract only the modulated components from the received signal. This effectively suppresses uncorrelated background noise and improves the SNR. The resulting demodulated signal is then processed using a Langevin-based model, which treats the waveform as a combination of deterministic signal and stochastic noise. This further isolates the scattering event with minimal distortion and no need for labeled data.

[0185] In this configuration, the lock-in amplifier serves as a narrowband filter centered at the modulation frequency, isolating relevant signal components while rejecting broadband noise. The output, containing primarily the scattered signal, is fed into the Langevin model, which uses a stochastic differential framework to iteratively suppress remaining noise. This layered approach combines real-time analog filtering with unsupervised, physics-informed de-noising to achieve high sensitivity in particle detection applications.

[0186] Example 11: Artificial Intelligence and Machine Learning Techniques and Approaches

[0187] Artificial intelligence is a broad field encompassing many techniques and approaches for creating machines that are able to perform prediction, pattern discovery, decision-making, reasoning, and other complex tasks normally associated with human intelligence [1], Machine learning is a sub-field of artificial intelligence wherein models are trained on input data to be able to make predictions based on new input data

[0001] [2], Machine learning comprises numerous methods that are applicable for implementing artificial intelligence enabled optical particle counters.Leydig Ref. 340550: 29-25 WO

[0188] Supervised Learning

[0189] Supervised learning entails providing a large dataset with inputs and known correct outputs, called labels. The model learns the relationship between the inputs and the output and after training is able to predict the output. Two main categories of supervised learning are regression models, which predict a numeric value, and classification models, which predict the category to which the input data belongs [2][3],

[0190] Many supervised machine learning algorithms may be used including linear regression, logistic regression, support vector machine (SVM), decision tree, random forest, neural network, and k-nearest neighbors. Detailed descriptions of these algorithms and examples of their usage are available in [1], [3], and

[0011] ,

[0191] Unsupervised Learning

[0192] In unsupervised learning, datasets provided for training the model contain inputs but not known correct outputs [2], The model learns patterns and groupings in the data instead of the relationship between the inputs and labels [3],

[0193] Clustering algorithms, the most common type of unsupervised learning, may be used and include centroid-based clustering, density-based clustering, distribution-based clustering, and hierarchical clustering [8],

[0194] Semi-supervised Learning

[0195] Semi-supervised learning first uses supervised learning on a small, labeled dataset for initial model training and then performs further training with unlabeled data [3], One approach, self-training, iteratively uses the model to infer predictions for the unlabeled data and then uses the outputs with the highest confidence to further train the model. This approach is useful when obtaining labeled data is inefficient yet unlabeled data is plentiful [1],

[0196] Data and TrainingLeydig Ref. 340550: 29-25 WO

[0197] Regardless of the type of machine learning leveraged, large amounts of data must be obtained for use in model training. The nature of the data is specific to the problem to which machine learning is being applied.

[0198] Model training may be performed offline or online. In offline training, a model is trained on a large amount of data at a single time, before it is incorporated into the device [4][5], The model may be infrequently retrained on more data, which requires an update of the implementation on the device. Offline training may be performed on a personal computer or a server, using a machine learning software tool such as TensorFlow or Scikit-learn [9]

[0010] , Online training involves continuously or frequently training and updating a model as the device collects data while in use in the field [6], This training approach requires either the training logic to be implemented on the device or training data to be transmitted off the device for processing and the resulting trained model transmitted back to the device [5],

[0199] Implementation

[0200] In order to use a trained model to make predictions on new input data, referred to as inference, the model must be implemented somewhere. A trained model may be implemented in several different ways. First, it may be implemented off of the device or on the device.

[0201] When a model is implemented off of the device, for example on a server, data is collected by the device and sent to the server for inference and the prediction is sent back to the device [3], This approach may be used when the model is too resource intensive for the device to compute.

[0202] The model may be implemented on the device if it has the necessary computational power. In this case, no external entity needs to be involved in data collection or inference. Simple to moderately complex models may be implemented as part of the embedded software that runs on the device processor or implemented in an FPGA or other programmable logic. Both approaches may utilize on-device machine learning frameworks and libraries to aid in implementation and computation [3], Implementation of more complex models that stretch the capabilities of theLeydig Ref. 340550: 29-25 WOdevice may be accomplished with hardware acceleration through FPGAs, ASICs, or GPUs [5],

[0203] Inference may also be dynamic or static. In dynamic inference, predictions are made on demand. This approach may be used when computation can be performed fast enough for the use case. For complex models in use cases that require immediate predictions, static inference may be used, where inference for a large set of common inputs is performed ahead of time and cached for efficient retrieval [7],

[0204] References relating to Example 11:

[0205] [1] developers.google.com / machine-learning / glossary

[0206] [2] developers.google.com / machine-learning / intro-to-ml / what-is-ml

[0207] [3] Jouini, 0.; Sethom, K.; Namoun, A.; Aljohani, N.; Alanazi, M.H.;Alanazi, M.N. A Survey of Machine Learning in Edge Computing: Techniques, Frameworks, Applications, Issues, and Research Directions. Technologies 2024, 12, 81. doi.org / 10.3390 / technologies12060081

[0208] [4] developers.google.com / machine-learning / crash-course / production-ml-systems / static-vs-dynamic-training

[0209] [5] Zhang, Z.; Li, J. A Review of Artificial Intelligence in Embedded Systems. Micromachines 2023, 14, 897. doi.org / 10.3390 / mi14050897

[0210] [6] Steven C.H. Hoi, Doyen Sahoo, Jing Lu, Peilin Zhao. Online learning: A comprehensive survey. Neurocomputing, Volume 459, 2021, Pages 249-289, ISSN 0925-2312, / doi.org / 10.1016 / j.neucom.2021.04.112

[0211] [7] developers.google.com / machine-learning / crash-course / production-ml-systems / static-vs-dynamic-inference

[0212] [8] developers.google.com / machine-learning / clustering / clustering-algorithms

[0213] [9] www.tensorflow.org / Leydig Ref. 340550: 29-25 WO

[0214]

[0010] scikit-learn.org / stable /

[0215]

[0011] Ajani, T.S.; Imoize, A.L.; Atayero, A. A. An Overview of Machine Learning within Embedded and Mobile Devices-Optimizations and Applications. Sensors 2021, 21, 4412. doi.org / 10.3390 / s21134412

[0216] Example 12: Sources of Training Data

[0217] Training data for the Al-enable particle detectors disclosed herein may be derived from multiple sources, including experimental or historical operational data collected from a single particle detector, experimental or historical operational data collected from multiple particle detectors, as well as synthetic training data. Synthetic training data is artificially generated information created by algorithms rather than collected from real-world particle detectors. In some aspects, synthetic training data may be designed to mimic the statistical patterns and characteristics of genuine data as closely as possible.

[0218] In other aspects, synthetic training datasets can be intentionally altered to include a synthetic bias. For example, a synthetic training data set can be altered to include an unusual amount of noise, or a particular signal pattern.

[0219] Example 13: Identifying Particle Composition and / or Source in a Cleanroom Environment via Particle Cluster Data and Cleanroom Operational Data

[0220] In one aspect, the composition and / or the source of particulate contamination in a cleanroom may be determined via analysis of particle count data from an optical particle counter combined with cleanroom operational data (e.g., programmable logic circuit input / output signals). For example, a seal inside an automated valve operating in an ultrapure water system, said water system serving a cleanroom, may begin to shed particulate as a result of wear. Turning now to FIGs.2 and 3, a water purification system is shown.

[0221] In the illustrated system, the particulate contamination from this deteriorating seal may cause a cluster or a series of clusters of particles of a particular size to be inadvertently released into the ultrapure water. The particles may then travel downstream, eventually reaching the optical particle counterLeydig Ref. 340550: 29-25 WOmonitoring the ultrapure water, and causing a series of spikes of particle count data in one or more size bins of the particle counter. The number of spikes in particle count data, the particular size bins associated with the spikes, as well as the time between the spikes may together form particle cluster data. In one aspect, an artificial intelligence model may overlay additional cleanroom operational data from the same time period (optionally corrected for transit time through the ultrapure water system before reaching the particle counter) allowing the artificial intelligence model to identify a particular signature of a process event in the ultrapure water system manifesting in the particle count data. In particular, the cleanroom operational data may include:• upstream process data such as temperature of the particle-containing fluid, pressure of the particle-containing fluid, flow rate of the particle containing fluid, pressure drop across a processing unit such as a filter;• equipment actuation signals such as process control input or output (I / O) signals from a process controller, the controller operating on the particlecontaining fluid, including:o output signals such as valve actuation signals or pump motor initiation signals sent from the process controller;o input signals such as valve limit switch signals being received by the process controller; and• electronic process maintenance data such as equipment-specific electronic sterilization records or electronic filter change records.

[0222] The artificial intelligence model may thus be able to identify a source (e.g., the deteriorating valve seat from a valve or group of valves) and / or a composition (e.g., polytetrafluoroethylene valve seat material) of the particulate contamination via analyzing a combination of particle cluster data including a series of spikes in particle counts in particular size bins occurring in recognizable pattern overtime combined with cleanroom operational data from the corresponding time period including valve actuation signals.

[0223] In one aspect, the artificial intelligence model may comprise a machine learning model configured to apply regression analysis to match particle cluster data to historical training data. The artificial intelligence model may be trained using anyLeydig Ref. 340550: 29-25 WOor all of the cleanroom operational data discussed above. Additionally, training the model may include correlating specific particle cluster data and cleanroom operational data signatures with particular process events. For example, in the valve seat deterioration example discussed above, the model may be trained to recognize a particle cluster data signature associated with valve seat deterioration in concert with a valve actuation signal (the upstream action causing the particle clusters). In one aspect, the connection between the particle cluster data and the cleanroom operational data may be established in training the model via particle contamination incident reports.

[0224] In one aspect, in response to identifying a source and / or composition of the particulate contamination via the particle cluster data and the cleanroom operational data, the artificial intelligence model may generate an output to a user. For example, the artificial intelligence model may generate a contamination alert to a user including the likely source of the particles (e.g., a deteriorating valve seat) and the likely composition of the particles (polytetrafluoroethylene).

[0225] Example 14: Particle Discrimination

[0226] In some embodiments, systems and methods of the invention combine on-axis particle measurements by detection of transmitted and forward scattered light, optionally using a differential detection configuration and / or using a structured beam such as a dark beam, with additional off-axis detection of scattered light or fluorescence for distinguishing between biological (e.g. microbial particles) and non-biological particles. Biological particles, such as microbes, are composed primarily of water, so they have a very low index of refraction contrast with water and hence scatter very little. Accordingly, biological particles generate a small side scattering signal or are not detected using side scattering. Biological particles do create a strong response signal in the present methods and systems using on-axis detection of transmitted light and forward scattered light, for example using structure beam and / or differential detection techniques. In some methods and systems of the invention, an artificial intelligence model may be configured to compare a large on-axis signal with the side scatter signal or a signal at the on-axis detector without a corresponding signal at the side scatter detector in order to characterize a particle asLeydig Ref. 340550: 29-25 WOa biological particle, such as a microbial particle. Alternatively, in some methods and systems of the invention, an artificial intelligence model may compare the on-axis signal and side scatter signal to characterize a particle as a non-biological particle.

[0227] In some embodiments, systems and methods of the invention use on-axis particle detection by detection of transmitted and forward scattered light using a structured beam, such as a dark beam, to characterize the refractive index of the particle, for example, relative to (e.g., greater or less than) the refractive index of the media the particle is in (e.g., the composition of the fluid flow). This aspect allows for determining attributes of the composition of the particle, for example, providing a means for distinguishing between metallic and non-metallic particles based on refractive index. In systems incorporating a differential detection configuration, for example, reliable and repeatable flipping of the “classical” particle signal may be used for accurately distinguishing between metallic and non-metallic particles. In some embodiments, for example, difference in the refractive index of the particle relative to the refractive index of the carrier fluid results in a signal that may be used by an artificial intelligence model to characterize the optical properties and composition of the particles.

[0228] If the analyzed particle has a refractive index higher than the refractive index of the carrier fluid, for example as in the case of polystyrene latex (PSL) particles in water media, a bright fringe is observed at the top of the beam and a dark fringe is observed at the bottom of the beam for conditions wherein the particle enters the beam in the flow cell. For example, when the particle enters the beam from the bottom. On the other hand, if the analyzed particle has a refractive index lower than the refractive index of the carrier fluid, as in the case of gold nanoparticles in water media, the fringe pattern is reversed for conditions of wherein the particle enters the beam in the flow cell, such as when the particle enters the beam from the bottom, such that a dark fringe is observed at the top of the beam and a bright fringe is observed at the bottom of the beam. Therefore, by observing and characterizing the sequence, order and / or positioning of the bright fringe and the dark fringe during the particle interaction with the beam, the refractive index relative to the carrier (e.g., greater than or less than) of the particles can be characterized by the artificialLeydig Ref. 340550: 29-25 WOintelligence model and, therefore, information relating to composition of the particles can also be inferred by the artificial intelligence model.

[0229] In another aspect, the artificial intelligence enabled particle characterization systems using differential detection as described herein may also characterize the refractive index of a particle as being higher or lower than the refractive index of the fluid. This is important, as metal particles will typically have a refractive index less than that of common fluids, while non-metals will have a refractive index higher than the fluid. Most metals are conductive and conductive materials are more harmful in many stages of the semiconductor manufacturing process, thus, distinguishing metals help identify more dangerous particles.

[0230] In an aspect, provided is an system for detecting particles in a fluid, the system comprising: i) a flow cell for flowing a liquid containing particles along a flow direction through a beam of electromagnetic radiation, ii) an optical source in optical communication with the flow cell for providing the beam of electromagnetic radiation, optionally a structured beam such as a dark beam; iii) a focusing system in optical communication with the optical source for focusing the beam of electromagnetic radiation into the flow cell; iv) an optical collection system for collecting and directing at least a portion of electromagnetic radiation onto a photodetector comprising at least two detector elements, wherein optionally the photodetector is positioned in optical communication with the flow cell for receiving incident electromagnetic radiation transmitted through the flow cell and electromagnetic radiation forwarded scattered by the particle; wherein each detector element produces an electric signal characteristic of the number and / or size of the particles detected and the photodetector characterizes the particles based on a differential signal generated from each detector element signal; wherein the detector characterizes the particles as having a lower or higher refractive index than the fluid. The detector may characterize, via an artificial intelligence model residing on the detector, the particle as a metal or a non-metal.

[0231] The photodetector may comprise at least two detector elements and characterizes the particles based on a differential signal from individual signals from each detector element indicative of the particles. The described systems mayLeydig Ref. 340550: 29-25 WOcomprise an artificial intelligence enabled analyzer for generating and / or analyzing the differential signal. The focusing system may direct the beam of electromagnetic radiation through the flow cell at least twice and the particles in the flow cell interact with a different portion of the beam on each individual pass through the flow cell. The analyzer may analyze the differential signal in the time domain.

[0232] Example 15: Combination of Measurements and Observations via an Artificial Intelligence Model

[0233] The quality of real-time measurements may be improved by expanding the training data to include laboratory analysis results of the same material which was measured in real time.

[0234] 1. RMM with microbial incubation results

[0235] Real-time microbial monitoring of fluids including water and air may be performed using a fluorescence particle counter, surface enhanced raman spectroscopy or other measurement techniques. Traditionally, microbial monitoring has been done by impacting an air sample onto a petri dish containing growth media or swabbing a liquid sample onto the growth media. The media is then cultured over a period of days and various types of microscopy analysis can be performed to determine the number and type of micro-organisms that were cultured. By using the laboratory results in the training data for machine learning based real time microbial monitoring, improved interpretation of the real-time measurements may be obtained.

[0236] 2. Nanoparticle optical particle counter vs sample capture / electron microscopy / sizing & composition

[0237] Real-time monitoring of micro and nano particles in fluids including liquids and air provides a measured concentration of particle contamination in clean fluids and it provides a time series of contamination trends. After passing through a particle counter, particles may be collected on a membrane surface for subsequent electron microscopy analysis including elemental composition. By incorporating the laboratory microscopy results as an input to the machine learning process, it may be possible to provide extended data interpretation. An example is, every time a certainLeydig Ref. 340550: 29-25 WOtime signature of varying particle concentration occurs, iron particles are dominant in the microscopy analysis.

[0238] Example 16: Al Agent

[0239] In one aspect an artificial intelligence enabled particle detector may comprise a large language model, operationally within the User Interface of the particle counter, which can provide advice to the user. The LLM is trained with data and operational scenarios (ex. Cleanroom certification or particle filter performance measurements) and technical interpretation of measurement data. This III agent can provide advice to the user about where to measure, how long they should measure, what do the measurements mean, etc. and have the ability to suggest alternative supporting measurements, depending on the task at hand.Leydig Ref. 340550: 29-25 WOASPECTSAspect 1. A method of detecting particles, the method comprising:passing a particle-containing fluid through a detection region of an optical particle detector;exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector;processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;determining a particle measurement output, wherein the particle measurement output is based at least in part on the artificial intelligence- enabled particle counter output.Aspect 2. The method of claim 1 , wherein the optical particle counter is for detecting particles corresponding to and / or originating from a cleanroom environment and / or controlled environment.Aspect 3. The method of claims 1 or 2, wherein the passing comprises flowing the particle-containing fluid through the detection region continuously or discontinuously.Aspect 4. The method of any of the preceding claims, wherein the photodetector comprises two or more photodetector elements or wherein the photodetector is a first photodetector and the sensor signal is a first sensor signal, the method further comprising generating a second sensor signal as a function of time via the second photodetector.Leydig Ref. 340550: 29-25 WOAspect 5. The method of any of the preceding claims, further comprising analyzing or archiving the particle measurement output or reporting the particle measurement output to a user.Aspect 6. The method of any of the preceding claims, wherein the sensor signal is one or more of a current signal, a voltage signal, and a pulse height signal from said photodetector as a function of time.Aspect 7. The method of any of the preceding claims, wherein the particle measurement output is a number of particle counts per time and / or per volume of fluid having a reduced number of false positive detection occurrences and / or having enhanced sensitivity for detection of particles having an average cross sectional dimension of less than or equal to 100 nm, optionally less than or equal to 50 nm, and optionally less than or equal to 20 nm.Aspect 8. The method of any of the preceding claims, further comprising processing system data as a function of time using said artificial intelligence model, thereby generating said artificial intelligence enabled particle counter output, wherein said system data corresponds to one or more of laser parameters (e.g. power, stability, condition), photodetector parameters (e.g., photodetector signal, biasing, noise, background, etc), calibration state parameters, optical bench parameters, or any combination of these.Aspect 9. The method of any of the preceding claims, wherein the processing comprises:segmenting the sensor signal, or the data set based thereon, into particle count data corresponding to a plurality of time intervals; andgenerating a particle detection probability map comprising a probability value for each time interval.Aspect 10. The method of any of the preceding claims, wherein the processing comprises:Leydig Ref. 340550: 29-25 WOsegmenting the sensor signal, or the data set based thereon, into particle count data corresponding to a plurality of time intervals, wherein the particle count data further includes a plurality of channels, each channel being correlated to a different particle size bin; andgenerating a particle detection probability map comprising a probability value for each time interval and each channel.Aspect 11. The method of any of the preceding claims, wherein the optical particle detector is a first optical particle detector and the sensor signal is a first sensor signal, the method further comprising:passing the particle-containing fluid through at least a second optical particle detector;producing a second sensor signal via the second optical particle detector;processing the second sensor signal, or a data set based thereon, via the artificial intelligence model, to generate the artificial intelligence enabled particle counter output.Aspect 12. The method of any of the preceding claims, comprising training the artificial intelligence model via synthetic data.Aspect 13. The method of any of the preceding claims, comprising training the artificial intelligence model via experimental and / or operational data.Aspect 14. The method of claim 11 , wherein the processing comprises comparing a first particle count derived from the first sensor signal to a second particle count derived from the second sensor signal, and wherein the artificial intelligence enabled particle counter output comprises an adjustment to a calibration parameter of the first and / or second optical particle counters.Aspect 15. The method of claim 14, wherein the adjustment to the calibration parameter of the first and / or second optical particle counters is configured toLeydig Ref. 340550: 29-25 WOharmonize a particle count of the first optical particle counter with a particle count of the second optical particle counter.Aspect 16. The method of any of the preceding claims, wherein the artificial intelligence model comprises a supervised, unsupervised, or semi-supervised regression model.Aspect 17. The method of any of the preceding claims, wherein the artificial intelligence model is trained with a training data set comprising features extracted from the sensor signal.Aspect 18. The method of claim 17, wherein the features comprise particle size, particle shape, particle speciation, the presence or absence of a particle, or any combination thereof.Aspect 19. The method of any of the preceding claims, wherein the sensor signal comprises unwanted noise, wherein the processing comprises filtering the unwanted noise via the artificial intelligence model, and wherein the artificial intelligence enabled particle counter output comprises filtered particle count data.Aspect 20. The method of claim 19, wherein the unwanted noise comprises a spike in the magnitude of the sensor signal that does not correspond to the presence of a particle in the detection region of the optical particle detector.Aspect 21. The method of claims 19 or 20, wherein the unwanted noise results from electronic noise, laser scattering anomalies, microbubble interference, and / or mechanical misalignment of the optical particle counter.Aspect 22. The method of any of the preceding claims, wherein the artificial intelligence model comprises a neural network or regression model.Aspect 23. The method of any one of the preceding claims wherein the artificial intelligence model is a machine learning model.Leydig Ref. 340550: 29-25 WOAspect 24. The method of any of the preceding claims, wherein processing the sensor signal comprises:counting a particle in response to identifying a sub-threshold particle detection event, wherein the identifying comprises comparing at least one characteristic of the sensor signal of the photodetector to one or more trained model characteristics of the artificial intelligence model.Aspect 25. The method of claim 24, wherein the sub-threshold particle detection event is defined as a peak in the magnitude of the sensor signal, the peak having a value below a predetermined particle detection threshold combined with the presence of a secondary indicator of a particle in the detection region.Aspect 26. The method of claim 25, wherein the secondary indicator comprises a pattern in the sensor signal.Aspect 27. The method of claim 26, wherein the pattern is one or more of pulse shape, pulse duration, pulse rise time, pulse fall time, a signal to noise ratio, or a frequency domain characteristic of the signal.Aspect 28. The method of claim 27, wherein the secondary indicator comprises a second signal from a second optical particle detector.Aspect 29. The method of any of the preceding claims, comprising:collecting historical data from the optical particle counter;analyzing the historical data; andpredicting, via the artificial intelligence model, a time in the future when maintenance or recalibration of the optical particle counter is needed.Aspect 30. The method of claim 29, comprising producing a maintenance alert to inform a user of the optical particle detector of the time in the future when maintenance or recalibration of the optical particle counter is needed.Leydig Ref. 340550: 29-25 WOAspect 31. The method of claims 29 or 30, comprising training the artificial intelligence model via a dataset comprising service records and test results.Aspect 32. The method of any of the preceding claims, wherein processing the sensor signal comprises characterizing, via the artificial intelligence model, the particles based on material composition and / or origin.Aspect 33. The method of claim 32, wherein the characterizing comprises distinguishing biological particles from non-biological particles.Aspect 34. The method of any of the preceding claims, wherein processing the sensor signal comprises comparing the sensor signal to historical signal data, and wherein the artificial intelligence enabled particle counter output comprises identifying a contamination event.Aspect 35. The method of any of the preceding claims, comprising receiving environmental data corresponding to ambient environmental conditions of the optical particle counter, and wherein processing the sensor signal comprises adjusting a particle count in response to a change in the ambient environmental conditions.Aspect 36. The method of any of the preceding claims, wherein the sensor signal comprises a first sensor signal produced from a first photodetector element and a second sensor signal produced via a second photodetector element, and wherein the processing comprises processing the first and second sensor signals to generate the artificial intelligence-enabled particle counter output.Aspect 37. A method of operating an optical particle counter, the method comprising:passing a particle-containing fluid through a detection region of an optical particle counter;generating system data via the optical particle counter and / or one or more sensors external to the optical particle counter;Leydig Ref. 340550: 29-25 WOprocessing the system data, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output; anddetermining an operational improvement output, wherein the operational improvement output is based at least in part on the artificial intelligence enabled particle counter output.Aspect 38. The method of claim 37, wherein the optical particle counter is for detecting particles corresponding to and / or originating from a cleanroom environment and / or controlled environment.Aspect 39. The method of claims 37 or 38, comprising:exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector; and detecting a particle in the detection region of the optical particle counter based on the sensor signal.Aspect 40. The method of any of claims 37-39, wherein the amount of light is an optical power or photon flux incident on the photodetector.Aspect 41. The method of any of claims 37-40, further comprising analyzing or archiving the operational improvement and / or reporting the operational improvement output to a user.Aspect 42. The method of any of claims 37-41 , wherein the receiving comprises receiving the system data from a data storage unit or sensor internal or external to the optical particle counter.Leydig Ref. 340550: 29-25 WOAspect 43. The method of any of claims 37-42, wherein the system data corresponds to one or more of laser parameters (e.g. power, stability, condition), photodetector parameters (e.g., photodetector signal, biasing, noise, background, etc), calibration state parameters, optical bench parameters, historical particle count data, or any combination of these.Aspect 44. The method of any of claims 37-43, wherein the system data corresponds to one or more of: pulse height signals from a photodetector of optical particle counter, a differential signal from a multi-element photodetector of the optical particle counter, or a signal shape.Aspect 45. The method of any of claims 37-44, wherein the optical particle counter is a first optical particle counter, and wherein the system data corresponds to a laser power, calibration state, alarm signal, photodetector stability and / or particle count data of a second optical particle counter.Aspect 46. The method of any of claims 37-45, wherein the system data comprises an individual unit historical alarm data set corresponding to stored particle count alarms of the optical particle counter.Aspect 47. The method of any of claims 37-46, wherein the optical particle counter is a first optical particle counter, and wherein the system data comprises a facility historical alarm data set corresponding to recorded particle count alarms of at least a second optical particle counter.Aspect 48. The method of any of claims 37-47, wherein the recorded particle count alarms of the second optical particle counter are derived from monitoring a particlecontaining fluid of the same cleanroom environment as the first optical particle counter.Leydig Ref. 340550: 29-25 WOAspect 49. The method of any of claims 37-48, wherein the recorded particle count alarms of the second optical particle counter are derived from monitoring the same particle-containing fluid as the first optical particle counter.Aspect 50. The method of any of claims 37-49, wherein the operational improvement output comprises an alarm probability map as a function of time.Aspect 51. The method of any of claims 37-50, wherein the operational improvement output comprises an alarm probability map as a function of time and location within the cleanroom environment.Aspect 52. The method of any of claims 37-51 , wherein the system data comprises a historical feature use data set corresponding to operational features of the optical particle counter used by a user.Aspect 53. The method of any of claims 37-52, wherein the operational improvement output comprises a change to a user interface layout and or workflow based at least in part on the historical feature use data set.Aspect 54. The method of any of claims 37-53, wherein the system data comprises a historical operational health data set corresponding to operational health parameters of the optical particle counter as a function of time.Aspect 55. The method of any of claims 37-54, wherein the operational health parameters include particle counts, particle alarms, false particle counts, sensor signal noise levels, and / or operational alarms of the optical particle counter.Aspect 56. The method of claims 54 or 55, wherein the operational improvement output comprises a suggested maintenance schedule for the optical particle counter.Aspect 57. The method of any of claims 37-56, wherein the system data comprises a historical particle count data set corresponding to recorded particle counts of the optical particle counter as a function of time.Leydig Ref. 340550: 29-25 WOAspect 58. The method of any of claims 23-57, wherein the operational improvement output comprises a particle count trend report relating to the optical particle counter.Aspect 59. The method of any of claims 37-58, wherein the system data comprises a cross-sensor data set corresponding to recorded particle information derived from a variety of sensors as a function of time.Aspect 60. The method of claim 59, wherein the sensors are selected from one or more particle detectors, molecular contamination detectors, and / or microbial detectors.Aspect 61. The method of claims 59 or 60, wherein the operational improvement output comprises a facility-wide contamination report relating to contamination levels of the particle-containing fluid.Aspect 62. The method of any of claims 37-61 , wherein the operational improvement output comprises a contamination event root case analysis report.Aspect 63. The method of any of claims 37-62, wherein the operational improvement output comprises a suggested alarm response report.Aspect 64. The method of any of claims 37-63, wherein the artificial intelligence model resides on a cloud-based database.Aspect 65. A method of detecting particles, the method comprising:passing a particle-containing fluid through a detection region of a particle detector;interrogating the fluid in the detection region, wherein the interrogation produces a sensor signal;processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;Leydig Ref. 340550: 29-25 WOdetermining a particle measurement output and / or an operational improvement output, based at least in part on the artificial intelligence-enabled particle counter output.Aspect 66. The method of claim 65, wherein the interrogating comprises exposing the fluid in the detection region to an acoustic wave.Aspect 67. The method of claim 65, wherein the interrogating comprises passing the fluid though a nanopore media, thereby producing an electric signal.Aspect 68. A method of classifying particles in a cleanroom, the method comprising:detecting particles in a fluid flowing in the cleanroom via an optical particle detector, the detecting comprising:flowing a particle-containing fluid through a detection region; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation generates scattered or emitted electromagnetic radiation from the particles;directing at least a portion of the scattered or emitted electromagnetic radiation from the particles onto a photodetector;generating particle count data via the photodetector, wherein the particle count data includes a plurality of channels, each channel being correlated to a different particle size bin;producing particle cluster data, the particle cluster data comprising particle count data for each size bin as a function of time;recording cleanroom operational data, the cleanroom operational data comprising:upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;upstream equipment actuation signals relating to processing of the particle-containing fluid upstream of the optical particle counter; orLeydig Ref. 340550: 29-25 WOelectronic process maintenance data relating to maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; andcorrelating, using an artificial intelligence model, the particle cluster data to the cleanroom operational data, and based thereon, determining a composition and / or source of the detected particles.Aspect 69. The method of claim 68, wherein the upstream process data comprises at least one of: a temperature of the particle-containing fluid, a pressure of the particle-containing fluid, a flow rate of the particle containing fluid, or pressure drop across a processing unit such as a filter, upstream of the optical particle counterAspect 70. The method claims 68 or 69, wherein the equipment actuation signals comprise process control input or output (I / O) signals from a process controller operating on the particle-containing fluid.Aspect 71. The method of claim 70, wherein the control input or output signals comprise at least of: a valve actuation signal sent from the process controller, a pump motor initiation signal sent from the process controller, or a valve limit switch signal received by the process controller.Aspect 72. The method of any of claims 68-71 , wherein the electronic process maintenance data comprises equipment-specific electronic sterilization records or electronic filter change records for one or more pieces of equipment operating on the particle-containing fluid upstream of the optical particle detector.Aspect 73. The method of any of claims 68-72, wherein the artificial intelligence model is a machine learning model configured to apply regression analysis to match particle cluster data to historical training data.Aspect 74. The method of claim 73, wherein the historical training data comprises one or more of:Leydig Ref. 340550: 29-25 WOhistorical upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;historical upstream equipment actuation signals relating to prior processing of the particle-containing fluid upstream of the optical particle counter;historical electronic process maintenance data relating to prior maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; orhistorical particle contamination incident reports relating to prior contamination events in the particle-containing fluid upstream of the optical particle counter.Aspect 75. The method of any of claims 68-74, wherein the particle cluster data comprises a plurality of particle count spikes in a least one size bin of the optical particle counter.Aspect 76. The method of any of claims 68-75 comprising generating, in response to determining the composition and / or the source of the detected particles, an output signal to a user.Aspect 77. The method of claim 76 wherein the output signal to the user comprises a contamination alert including the likely source of the particles or the likely composition of the particles.Aspect 78. A system for detecting particles, the system comprising:an optical particle detector configured for:passing a particle-containing fluid through a detection region thereof; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;Leydig Ref. 340550: 29-25 WOgenerating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector; andan artificial intelligence model configured for:processing the sensor signal, or a data set based thereon to generate an artificial intelligence-enabled particle counter output; anddetermining a particle measurement output, wherein the particle measurement output is based at least in part on the artificial intelligence- enabled particle counter output.Aspect 79. A system for detecting particles, the system comprising:an optical particle counter configured for:passing a particle-containing fluid through a detection region thereof; generating system data via the optical particle counter and / or one or more sensors external to the optical particle counter; andan artificial intelligence model configured for:processing the system data to generate an artificial intelligence- enabled particle counter output; anddetermining an operational improvement output, wherein the operational improvement output is based at least in part on the artificial intelligence enabled particle counter output.Aspect 80. A system for detecting particles, the system comprising:an optical particle counter configured for:passing a particle-containing fluid through a detection region of the particle detector;interrogating the fluid in the detection region, wherein the interrogation produces a sensor signal; andan artificial intelligence model configured for:processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;Leydig Ref. 340550: 29-25 WOdetermining a particle measurement output and / or an operational improvement output, based at least in part on the artificial intelligence-enabled particle counter output.Aspect 81. A system for classifying particles in a cleanroom, the system comprising:an optical particle counter configured for:detecting particles in a fluid flowing in the cleanroom via an optical particle detector, the detecting comprising:flowing a particle-containing fluid through a detection region; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation generates scattered or emitted electromagnetic radiation from the particles;directing at least a portion of the scattered or emitted electromagnetic radiation from the particles onto a photodetector;generating particle count data via the photodetector, wherein the particle count data includes a plurality of channels, each channel being correlated to a different particle size bin; andan artificial intelligence model configured for:producing particle cluster data, the particle cluster data comprising particle count data for each size bin as a function of time;recording cleanroom operational data, the cleanroom operational data comprising:upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;upstream equipment actuation signals relating to processing of the particle-containing fluid upstream of the optical particle counter; or electronic process maintenance data relating to maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; andLeydig Ref. 340550: 29-25 WOcorrelating, using the artificial intelligence model, the particle cluster data to the cleanroom operational data, and based thereon, determining a composition and / or source of the detected particles.STATEMENTS REGARDING INCORPORATION BY REFERENCE AND VARIATIONS

[0240] All references throughout this application, for example patent documents including issued or granted patents or equivalents; patent application publications; and non-patent literature documents or other source material; are hereby incorporated by reference herein in their entireties, as though individually incorporated by reference, to the extent each reference is at least partially not inconsistent with the disclosure in this application (for example, a reference that is partially inconsistent is incorporated by reference except for the partially inconsistent portion of the reference).

[0241] The terms and expressions which have been employed herein are used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions of excluding any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention claimed. Thus, it should be understood that although the present invention has been specifically disclosed by preferred embodiments, exemplary embodiments and optional features, modification and variation of the concepts herein disclosed may be resorted to by those skilled in the art, and that such modifications and variations are considered to be within the scope of this invention as defined by the appended claims. The specific embodiments provided herein are examples of useful embodiments of the present invention and it will be apparent to one skilled in the art that the present invention may be carried out using a large number of variations of the devices, device components, methods steps set forth in the present description. As will be obvious to one of skill in the art, methods and devices useful for the present methods can include a large number of optional composition and processing elements and steps.

[0242] As used herein and in the appended claims, the singular forms "a", "an", and "the" include plural reference unless the context clearly dictates otherwise.Thus, for example, reference to "a cell" includes a plurality of such cells andLeydig Ref. 340550: 29-25 WOequivalents thereof known to those skilled in the art. As well, the terms "a" (or "an"), "one or more" and "at least one" can be used interchangeably herein. It is also to be noted that the terms "comprising", "including", and "having" can be used interchangeably. The expression “of any of claims XX-YY” (wherein XX and YY refer to claim numbers) is intended to provide a multiple dependent claim in the alternative form, and in some embodiments is interchangeable with the expression “as in any one of claims XX-YY.”

[0243] When a Markush group or other grouping is used herein, all individual members of the group and all combinations and subcombinations possible of the group are intended to be individually included in the disclosure.

[0244] Every device, system, formulation, combination of components, or method described or exemplified herein can be used to practice the invention, unless otherwise stated.

[0245] Whenever a range is given in the specification, for example, a temperature range, a time range, or a composition or concentration range, all intermediate ranges and subranges, as well as all individual values included in the ranges given are intended to be included in the disclosure. It will be understood that any subranges or individual values in a range or subrange that are included in the description herein can be excluded from the claims herein.

[0246] All patents and publications mentioned in the specification are indicative of the levels of skill of those skilled in the art to which the invention pertains.References cited herein are incorporated by reference herein in their entirety to indicate the state of the art as of their publication or filing date and it is intended that this information can be employed herein, if needed, to exclude specific embodiments that are in the prior art. For example, when composition of matter are claimed, it should be understood that compounds known and available in the art prior to Applicant's invention, including compounds for which an enabling disclosure is provided in the references cited herein, are not intended to be included in the composition of matter claims herein.Leydig Ref. 340550: 29-25 WO

[0247] As used herein, “comprising” is synonymous with "including," "containing," or "characterized by," and is inclusive or open-ended and does not exclude additional, unrecited elements or method steps. As used herein, "consisting of" excludes any element, step, or ingredient not specified in the claim element. As used herein, "consisting essentially of" does not exclude materials or steps that do not materially affect the basic and novel characteristics of the claim. In each instance herein any of the terms "comprising", "consisting essentially of" and "consisting of" may be replaced with either of the other two terms. The invention illustratively described herein suitably may be practiced in the absence of any element or elements, limitation or limitations which is not specifically disclosed herein.

[0248] One of ordinary skill in the art will appreciate that starting materials, biological materials, reagents, synthetic methods, purification methods, analytical methods, assay methods, and biological methods other than those specifically exemplified can be employed in the practice of the invention without resort to undue experimentation. All art-known functional equivalents, of any such materials and methods are intended to be included in this invention. The terms and expressions which have been employed are used as terms of description and not of limitation, and there is no intention that in the use of such terms and expressions of excluding any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention claimed. Thus, it should be understood that although the present invention has been specifically disclosed by preferred embodiments and optional features, modification and variation of the concepts herein disclosed may be resorted to by those skilled in the art, and that such modifications and variations are considered to be within the scope of this invention as defined by the appended claims.

[0249] While certain features and aspects have been described with respect to exemplary embodiments, one skilled in the art will recognize that numerous modifications are possible. For example, the methods and processes described herein may be implemented using hardware components, software components, and / or any combination thereof. Further, while various methods and processes described herein may be described with respect to particular structural and / or functional components for ease of description, methods provided by variousLeydig Ref. 340550: 29-25 WOembodiments are not limited to any particular structural and / or functional architecture but instead can be implemented on any suitable hardware, firmware and / or software configuration. Similarly, while certain functionality is ascribed to certain system components, unless the context dictates otherwise, this functionality can be distributed among various other system components in accordance with the several embodiments.

[0250] Moreover, while the procedures of the methods and processes described herein are described in a particular order for ease of description, unless the context dictates otherwise, various procedures may be reordered, added, and / or omitted in accordance with various embodiments. Moreover, the procedures described with respect to one method or process may be incorporated within other described methods or processes; likewise, system components described according to a particular structural architecture and / or with respect to one system may be organized in alternative structural architectures and / or incorporated within other described systems. Hence, while various embodiments are described with — or without — certain features for ease of description and to illustrate exemplary aspects of those embodiments, the various components and / or features described herein with respect to a particular embodiment can be substituted, added and / or subtracted from among other described embodiments, unless the context dictates otherwise. Consequently, although several exemplary embodiments are described above, it will be appreciated that the invention is intended to cover all modifications and equivalents within the scope of the following claims.

Claims

Leydig Ref. 340550: 29-25 WOWe claim:

1. A method of detecting particles, the method comprising:passing a particle-containing fluid through a detection region of an optical particle detector;exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector;processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;determining a particle measurement output, wherein the particle measurement output is based at least in part on the artificial intelligence- enabled particle counter output.

2. The method of claim 1 , wherein the optical particle counter is for detecting particles corresponding to and / or originating from a cleanroom environment and / or controlled environment.

3. The method of claims 1 or 2, wherein the passing comprises flowing the particle-containing fluid through the detection region continuously or discontinuously.

4. The method of any of the preceding claims, wherein the photodetector comprises two or more photodetector elements or wherein the photodetector is a first photodetector and the sensor signal is a first sensor signal, the method further comprising generating a second sensor signal as a function of time via the second photodetector.Leydig Ref. 340550: 29-25 WO5. The method of any of the preceding claims, further comprising analyzing or archiving the particle measurement output or reporting the particle measurement output to a user.

6. The method of any of the preceding claims, wherein the sensor signal is one or more of a current signal, a voltage signal, and a pulse height signal from said photodetector as a function of time.

7. The method of any of the preceding claims, wherein the particle measurement output is a number of particle counts per time and / or per volume of fluid having a reduced number of false positive detection occurrences and / or having enhanced sensitivity for detection of particles having an average cross sectional dimension of less than or equal to 100 nm, optionally less than or equal to 50 nm, and optionally less than or equal to 20 nm.

8. The method of any of the preceding claims, further comprising processing system data as a function of time using said artificial intelligence model, thereby generating said artificial intelligence enabled particle counter output, wherein said system data corresponds to one or more of laser parameters, photodetector parameters, calibration state parameters, optical bench parameters, or any combination of these.

9. The method of any of the preceding claims, wherein the processing comprises:segmenting the sensor signal, or the data set based thereon, into particle count data corresponding to a plurality of time intervals; andgenerating a particle detection probability map comprising a probability value for each time interval.

10. The method of any of the preceding claims, wherein the processingcomprises:Leydig Ref. 340550: 29-25 WOsegmenting the sensor signal, or the data set based thereon, into particle count data corresponding to a plurality of time intervals, wherein the particle count data further includes a plurality of channels, each channel being correlated to a different particle size bin; andgenerating a particle detection probability map comprising a probability value for each time interval and each channel.

11. The method of any of the preceding claims, wherein the optical particle detector is a first optical particle detector and the sensor signal is a first sensor signal, the method further comprising:passing the particle-containing fluid through at least a second optical particle detector;producing a second sensor signal via the second optical particle detector;processing the second sensor signal, or a data set based thereon, via the artificial intelligence model, to generate the artificial intelligence enabled particle counter output.

12. The method of any of the preceding claims, comprising training the artificial intelligence model via synthetic data.

13. The method of any of the preceding claims, comprising training the artificial intelligence model via experimental and / or operational data.

14. The method of claim 11 , wherein the processing comprises comparing a first particle count derived from the first sensor signal to a second particle count derived from the second sensor signal, and wherein the artificial intelligence enabled particle counter output comprises an adjustment to a calibration parameter of the first and / or second optical particle counters.

15. The method of claim 14, wherein the adjustment to the calibration parameter of the first and / or second optical particle counters is configured to harmonize aLeydig Ref. 340550: 29-25 WOparticle count of the first optical particle counter with a particle count of the second optical particle counter.

16. The method of any of the preceding claims, wherein the artificial intelligence model comprises a supervised, unsupervised, or semi-supervised regression model.

17. The method of any of the preceding claims, wherein the artificial intelligence model is trained with a training data set comprising features extracted from the sensor signal.

18. The method of claim 17, wherein the features comprise particle size, particle shape, particle speciation, the presence or absence of a particle, or any combination thereof.

19. The method of any of the preceding claims, wherein the sensor signal comprises unwanted noise, wherein the processing comprises filtering the unwanted noise via the artificial intelligence model, and wherein the artificial intelligence enabled particle counter output comprises filtered particle count data.

20. The method of claim 19, wherein the unwanted noise comprises a spike in the magnitude of the sensor signal that does not correspond to the presence of a particle in the detection region of the optical particle detector.

21. The method of claims 19 or 20, wherein the unwanted noise results from electronic noise, laser scattering anomalies, microbubble interference, and / or mechanical misalignment of the optical particle counter.

22. The method of any of the preceding claims, wherein the artificial intelligence model comprises a neural network or regression model.

23. The method of any one of the preceding claims wherein the artificial intelligence model is a machine learning model.Leydig Ref. 340550: 29-25 WO24. The method of any of the preceding claims, wherein processing the sensor signal comprises:counting a particle in response to identifying a sub-threshold particle detection event, wherein the identifying comprises comparing at least one characteristic of the sensor signal of the photodetector to one or more trained model characteristics of the artificial intelligence model.

25. The method of claim 24, wherein the sub-threshold particle detection event is defined as a peak in the magnitude of the sensor signal, the peak having a value below a predetermined particle detection threshold combined with the presence of a secondary indicator of a particle in the detection region.

26. The method of claim 25, wherein the secondary indicator comprises a pattern in the sensor signal.

27. The method of claim 26, wherein the pattern is one or more of pulse shape, pulse duration, pulse rise time, pulse fall time, a signal to noise ratio, or a frequency domain characteristic of the signal.

28. The method of claim 27, wherein the secondary indicator comprises a second signal from a second optical particle detector.

29. The method of any of the preceding claims, comprising:collecting historical data from the optical particle counter;analyzing the historical data; andpredicting, via the artificial intelligence model, a time in the future when maintenance or recalibration of the optical particle counter is needed.

30. The method of claim 29, comprising producing a maintenance alert to inform a user of the optical particle detector of the time in the future when maintenance or recalibration of the optical particle counter is needed.Leydig Ref. 340550: 29-25 WO31. The method of claims 29 or 30, comprising training the artificial intelligence model via a dataset comprising service records and test results.

32. The method of any of the preceding claims, wherein processing the sensor signal comprises characterizing, via the artificial intelligence model, the particles based on material composition and / or origin.

33. The method of claim 32, wherein the characterizing comprises distinguishing biological particles from non-biological particles.

34. The method of any of the preceding claims, wherein processing the sensor signal comprises comparing the sensor signal to historical signal data, and wherein the artificial intelligence enabled particle counter output comprises identifying a contamination event.

35. The method of any of the preceding claims, comprising receiving environmental data corresponding to ambient environmental conditions of the optical particle counter, and wherein processing the sensor signal comprises adjusting a particle count in response to a change in the ambient environmental conditions.

36. The method of any of the preceding claims, wherein the sensor signal comprises a first sensor signal produced from a first photodetector element and a second sensor signal produced via a second photodetector element, and wherein the processing comprises processing the first and second sensor signals to generate the artificial intelligence-enabled particle counter output.

37. A method of operating an optical particle counter, the method comprising:passing a particle-containing fluid through a detection region of an optical particle counter;generating system data via the optical particle counter and / or one or more sensors external to the optical particle counter;Leydig Ref. 340550: 29-25 WOprocessing the system data, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output; anddetermining an operational improvement output, wherein the operational improvement output is based at least in part on the artificial intelligence enabled particle counter output.

38. The method of claim 37, wherein the optical particle counter is for detecting particles corresponding to and / or originating from a cleanroom environment and / or controlled environment.

39. The method of claims 37 or 38, comprising:exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;generating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector; and detecting a particle in the detection region of the optical particle counter based on the sensor signal.

40. The method of any of claims 37-39, wherein the amount of light is an optical power or photon flux incident on the photodetector.

41. The method of any of claims 37-40, further comprising analyzing or archiving the operational improvement and / or reporting the operational improvement output to a user.

42. The method of any of claims 37-41 , wherein the receiving comprises receiving the system data from a data storage unit or sensor internal or external to the optical particle counter.Leydig Ref. 340550: 29-25 WO43. The method of any of claims 37-42, wherein the system data corresponds to one or more of laser parameters (e.g. power, stability, condition), photodetector parameters (e.g., photodetector signal, biasing, noise, background, etc), calibration state parameters, optical bench parameters, historical particle count data, or any combination of these.

44. The method of any of claims 37-43, wherein the system data corresponds to one or more of: pulse height signals from a photodetector of optical particle counter, a differential signal from a multi-element photodetector of the optical particle counter, or a signal shape.

45. The method of any of claims 37-44, wherein the optical particle counter is a first optical particle counter, and wherein the system data corresponds to a laser power, calibration state, alarm signal, photodetector stability and / or particle count data of a second optical particle counter.

46. The method of any of claims 37-45, wherein the system data comprises an individual unit historical alarm data set corresponding to stored particle count alarms of the optical particle counter.

47. The method of any of claims 37-46, wherein the optical particle counter is a first optical particle counter, and wherein the system data comprises a facility historical alarm data set corresponding to recorded particle count alarms of at least a second optical particle counter.

48. The method of any of claims 37-47, wherein the recorded particle count alarms of the second optical particle counter are derived from monitoring a particlecontaining fluid of the same cleanroom environment as the first optical particle counter.Leydig Ref. 340550: 29-25 WO49. The method of any of claims 37-48, wherein the recorded particle count alarms of the second optical particle counter are derived from monitoring the same particle-containing fluid as the first optical particle counter.

50. The method of any of claims 37-49, wherein the operational improvement output comprises an alarm probability map as a function of time.

51. The method of any of claims 37-50, wherein the operational improvement output comprises an alarm probability map as a function of time and location within the cleanroom environment.

52. The method of any of claims 37-51 , wherein the system data comprises a historical feature use data set corresponding to operational features of the optical particle counter used by a user.

53. The method of any of claims 37-52, wherein the operational improvement output comprises a change to a user interface layout and or workflow based at least in part on the historical feature use data set.

54. The method of any of claims 37-53, wherein the system data comprises a historical operational health data set corresponding to operational health parameters of the optical particle counter as a function of time.

55. The method of any of claims 37-54, wherein the operational health parameters include particle counts, particle alarms, false particle counts, sensor signal noise levels, and / or operational alarms of the optical particle counter.

56. The method of claims 54 or 55, wherein the operational improvement output comprises a suggested maintenance schedule for the optical particle counter.

57. The method of any of claims 37-56, wherein the system data comprises a historical particle count data set corresponding to recorded particle counts of the optical particle counter as a function of time.Leydig Ref. 340550: 29-25 WO58. The method of any of claims 23-57, wherein the operational improvement output comprises a particle count trend report relating to the optical particle counter.

59. The method of any of claims 37-58, wherein the system data comprises a cross-sensor data set corresponding to recorded particle information derived from a variety of sensors as a function of time.

60. The method of claim 59, wherein the sensors are selected from one or more particle detectors, molecular contamination detectors, and / or microbial detectors.

61. The method of claims 59 or 60, wherein the operational improvement output comprises a facility-wide contamination report relating to contamination levels of the particle-containing fluid.

62. The method of any of claims 37-61 , wherein the operational improvement output comprises a contamination event root case analysis report.

63. The method of any of claims 37-62, wherein the operational improvement output comprises a suggested alarm response report.

64. The method of any of claims 37-63, wherein the artificial intelligence model resides on a cloud-based database.

65. A method of detecting particles, the method comprising:passing a particle-containing fluid through a detection region of a particle detector;interrogating the fluid in the detection region, wherein the interrogation produces a sensor signal;processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;Leydig Ref. 340550: 29-25 WOdetermining a particle measurement output and / or an operational improvement output, based at least in part on the artificial intelligence-enabled particle counter output.

66. The method of claim 65, wherein the interrogating comprises exposing the fluid in the detection region to an acoustic wave.

67. The method of claim 65, wherein the interrogating comprises passing the fluid though a nanopore media, thereby producing an electric signal.

68. A method of classifying particles in a cleanroom, the method comprising:detecting particles in a fluid flowing in the cleanroom via an optical particle detector, the detecting comprising:flowing a particle-containing fluid through a detection region; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation generates scattered or emitted electromagnetic radiation from the particles;directing at least a portion of the scattered or emitted electromagnetic radiation from the particles onto a photodetector;generating particle count data via the photodetector, wherein the particle count data includes a plurality of channels, each channel being correlated to a different particle size bin;producing particle cluster data, the particle cluster data comprising particle count data for each size bin as a function of time;recording cleanroom operational data, the cleanroom operational data comprising:upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;upstream equipment actuation signals relating to processing of the particle-containing fluid upstream of the optical particle counter; orLeydig Ref. 340550: 29-25 WOelectronic process maintenance data relating to maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; andcorrelating, using an artificial intelligence model, the particle cluster data to the cleanroom operational data, and based thereon, determining a composition and / or source of the detected particles.

69. The method of claim 68, wherein the upstream process data comprises at least one of: a temperature of the particle-containing fluid, a pressure of the particlecontaining fluid, a flow rate of the particle containing fluid, or pressure drop across a processing unit such as a filter, upstream of the optical particle counter.

70. The method claims 68 or 69, wherein the equipment actuation signals comprise process control input or output (I / O) signals from a process controller operating on the particle-containing fluid.

71. The method of claim 70, wherein the control input or output signals comprise at least of: a valve actuation signal sent from the process controller, a pump motor initiation signal sent from the process controller, or a valve limit switch signal received by the process controller.

72. The method of any of claims 68-71 , wherein the electronic process maintenance data comprises equipment-specific electronic sterilization records or electronic filter change records for one or more pieces of equipment operating on the particle-containing fluid upstream of the optical particle detector.

73. The method of any of claims 68-72, wherein the artificial intelligence model is a machine learning model configured to apply regression analysis to match particle cluster data to historical training data.

74. The method of claim 73, wherein the historical training data comprises one or more of:Leydig Ref. 340550: 29-25 WOhistorical upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;historical upstream equipment actuation signals relating to prior processing of the particle-containing fluid upstream of the optical particle counter;historical electronic process maintenance data relating to prior maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; orhistorical particle contamination incident reports relating to prior contamination events in the particle-containing fluid upstream of the optical particle counter.

75. The method of any of claims 68-74, wherein the particle cluster data comprises a plurality of particle count spikes in a least one size bin of the optical particle counter.

76. The method of any of claims 68-75 comprising generating, in response to determining the composition and / or the source of the detected particles, an output signal to a user.

77. The method of claim 76 wherein the output signal to the user comprises a contamination alert including the likely source of the particles or the likely composition of the particles.

78. A system for detecting particles, the system comprising:an optical particle detector configured for:passing a particle-containing fluid through a detection region thereof; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation results in scattered, emitted or transmitted electromagnetic radiation from the particles;directing at least a portion of the scattered, emitted, or transmitted electromagnetic radiation from the particles onto a photodetector;Leydig Ref. 340550: 29-25 WOgenerating a sensor signal as a function of time via the photodetector, the sensor signal corresponding to an amount of light incident on the photodetector; andan artificial intelligence model configured for:processing the sensor signal, or a data set based thereon to generate an artificial intelligence-enabled particle counter output; anddetermining a particle measurement output, wherein the particle measurement output is based at least in part on the artificial intelligence- enabled particle counter output.

79. A system for detecting particles, the system comprising:an optical particle counter configured for:passing a particle-containing fluid through a detection region thereof; generating system data via the optical particle counter and / or one or more sensors external to the optical particle counter; andan artificial intelligence model configured for:processing the system data to generate an artificial intelligence- enabled particle counter output; anddetermining an operational improvement output, wherein the operational improvement output is based at least in part on the artificial intelligence enabled particle counter output.

80. A system for detecting particles, the system comprising:an optical particle counter configured for:passing a particle-containing fluid through a detection region of the particle detector;interrogating the fluid in the detection region, wherein the interrogation produces a sensor signal; andan artificial intelligence model configured for:processing the sensor signal, or a data set based thereon, using an artificial intelligence model and / or output from an artificial intelligence model, to generate an artificial intelligence-enabled particle counter output;Leydig Ref. 340550: 29-25 WOdetermining a particle measurement output and / or an operational improvement output, based at least in part on the artificial intelligence-enabled particle counter output.

81. A system for classifying particles in a cleanroom, the system comprising: an optical particle counter configured for:detecting particles in a fluid flowing in the cleanroom via an optical particle detector, the detecting comprising:flowing a particle-containing fluid through a detection region; exposing the fluid in the detection region to a beam of electromagnetic radiation, wherein interaction between particles of the particle-containing fluid and the beam of electromagnetic radiation generates scattered or emitted electromagnetic radiation from the particles;directing at least a portion of the scattered or emitted electromagnetic radiation from the particles onto a photodetector;generating particle count data via the photodetector, wherein the particle count data includes a plurality of channels, each channel being correlated to a different particle size bin; andan artificial intelligence model configured for:producing particle cluster data, the particle cluster data comprising particle count data for each size bin as a function of time;recording cleanroom operational data, the cleanroom operational data comprising:upstream process data relating to processing of the particlecontaining fluid upstream of the optical particle counter;upstream equipment actuation signals relating to processing of the particle-containing fluid upstream of the optical particle counter; or electronic process maintenance data relating to maintenance or sterilization of equipment processing the particle-containing fluid upstream of the optical particle counter; andLeydig Ref. 340550: 29-25 WOcorrelating, using the artificial intelligence model, the particle cluster data to the cleanroom operational data, and based thereon, determining a composition and / or source of the detected particles.