Metadynamics sampling in atomic environment space for collecting training data for machine learning potentials

The universal mathematical form of machine-learning potentials (MLPs) shifts the core of development of interatomic potentials to collecting proper training data. Ideally, the training set should encompass diverse local atomic environments but conventional approaches are prone to sampling similar configurations repeatedly, mainly due to the Boltzmann statistics. As such, practitioners handpick a large pool of distinct configurations manually, stretching the development period significantly. To overcome this hurdle, methods are being proposed that automatically generate training data. Herein, we suggest a sampling method optimized for gathering diverse yet relevant configurations semi-automatically. This is achieved by applying the metadynamics with the descriptor for the local atomic environment as a collective variable. As a result, the simulation is automatically steered toward unvisited local environment space such that each atom experiences diverse chemical environments without redundancy. We apply the proposed metadynamics sampling to H:Pt(111), GeTe, and Si systems. Throughout these examples, a small number of metadynamics trajectories can provide reference structures necessary for training high-fidelity MLPs. By proposing a semi-automatic sampling method tuned for MLPs, the present work paves the way to wider applications of MLPs to many challenging applications.

Realizing Value from Adherence to Crude Oil Custody Transfer Automatic Sampling Standards

The successful implementation of a crude oil custody transfer sampling system is a key component to achieving a desired measurement uncertainty for a crude oil metering station or loading/unloading point. Our analysis of thousands of installations worldwide provides practical examples of how operators can be confident that when they install a sampling system, it will deliver the uncertainty that they need to meet the overall custody transfer requirements. Crude oil sampling for custody transfer becomes more challenging as production flow rates decline, oil fields mature, and water cut content increases. It is therefore important that the performance of a sampling system is evaluated on a regular basis and that any limitations are identified. Any risk of change in performance or measurement uncertainty can then be prioritized or minimized. International standards and contracts determine the allowable uncertainty for net oil (oil minus water) for custody transfer/allocation. For accurate allocation of the sources of crude oil or the application of tax tariffs, fluids must be measured prior to being comingled. Automatic crude oil sampling can be challenging because it can require high-energy mixing with low power consumption and negligible pressure loss to overcome stratification and allow representative sampling. The certification, or "proving", of the sampling system provides a basis for establishing and verifying the system's true uncertainty at worst case conditions. There is an abundance of computational fluid dynamics (CFD) simulations and studies on crude oil (and water) mixing. However, these are abstract because of the uncertainty in where the water is located and how it may be dispersed at the boundary conditions of any simulation. To provide more robust simulations, we expanded on the established methods by combining simulation data with known theoretical calculations and engineering laboratory test data as well as hundreds of certifications (proving) results from around the world. Automatic sampling systems using dynamic mixing technology delivers a unique solution that enables operators to minimize the quality measurement uncertainty, improve overall balance, and reduce financial loss (and unaccounted for) in custody transfer quality measurements.

Beyond high frequency monitoring: an optimised automatic sampling

<p>Beyond high frequency monitoring : an optimised automatic sampling</p><p>Mougin Jérémy, Superville Pierre-Jean, Cornard Jean-Paul, Billon Gabriel</p><p> </p><p>In order to improve the representativity of samples when monitoring a water body, efforts have been made these last years to develop new methodologies to replace grab samples. Passive samplers have allowed to have measurement averaged over several days and represented a first step. High frequency monitoring (usually one measure per hour), either in situ or on-line, led to the observations of daily cycles or transitory phenomena that were not suspected beforehand.</p><p>However, such method is usually difficult to implement for some trace analytes (e.g. trace metals or pesticides) or for some specific analysis (e.g. size exclusion chromatography on natural organic matter). Automatic sampling and analysis in the lab can be a solution, but it becomes very labor intensive as soon as the sampling frequency is high. Luck is also needed as a long sampling period can sometimes lead to very few variations if the water system is stable. In order to optimise the automatic sampling, a new methodology has been developped in this project.</p><p>A multiparameter probe measuring general parameters (temperature, pH, turbidity, ORP, conductivity, dissolved oxygen and two fluorometers for organic matter) was coupled with an automatic filtering sampler. The data from the probe are processed on-line and an algorithm decides if the geochemical situation in the water body seems new enough to trigger the sampling, based on previously sampled waters. The aim of this device is to collect the right number of samples with the best representativeness of phenomena taking place in the environment.</p><p>This method will be tested over a year in 2021 in order to monitor the dissolved organic matter in a small stream with both rural and urban contamination. These high-frequency measurements and samplings could make it possible to better define the sources and dynamics of the organic matter that has a strong impact on the quality of watercourses.</p>

Modeling annual diffuse phosphorus concentrations in Danish mini-catchments

<p>Intensive monitoring data from the Danish National Monitoring programme (NOVANA) from 24 smaller catchments (mean: 14 km<sup>2</sup>) was used in a two layer cross-validation to establish a model for the annual diffuse phosphorus (P) flow-weighted concentration in Danish streams. A total of 196 monitoring years with data from automatic sampling (ISCO) of water from the 24 streams were used as a training dataset. Data in the training dataset covers the period 1994-2002.</p><p>Moreover, another dataset consisting of 108 agricultural mini-catchments with discrete water samples covering the period 1990-2017 was used as a control in the eight different georegions of Denmark. A total of four different models was established three models based on the intensive dataset and one model based on the larger dataset with discrete water sampling.</p><p>The best model established included eight explanatory parameters and explained 53 % of the variation in the annual flow-weighted total P concentrations in the training dataset. A validation of the four different models established showed that the best model has to be bias-corrected in some of the georegions. The result of the validation shows that the models generally overestimate the total P concentrations.   An overestimation of around 10-20% was to be expected as intensive automatic water sampling in streams has shown that the flow-weighted concentration of total P obtained from discrete sampling (monthly or fortnightly) is normally underestimated.</p><p>The validations of the three models based on intensive dataset showed an R-square between 0.08 and 0.12. The model based on the larger data with discrete samples had an R-square (0.29).</p>

Modifying an ÄKTApurifier System for the Automated Acquisition of Samples for Kinetic Modeling of Batch Reactions

Recording the data necessary to assess the kinetics of a reaction can be labor-intensive. In this technology brief, we show a method to automate this task by utilizing parts of an ÄKTApurifier chromatography system to automatically take samples from a reaction vessel at predefined time intervals and place them in 96-well plates and also enable correlating the samples with in-line spectral data of the reaction solution. Automatic sampling can reduce experimental bottlenecks by enabling overnight reactions or a higher degree of parallelization. To demonstrate the feasibility of the method, we performed batch-PEGylation of lysozyme with varying conditions by changing the molar excess of the PEG reagent. We used analytical cation-exchange chromatography to analyze the samples taken during the batch reaction, determining the concentrations of the individual species present at each time step. Subsequently, we fitted a kinetic model on these data. Fitting the model to four different reaction conditions simultaneously yielded a regression coefficient of R2 = 0.871.

Systems of grained materials automated sampling from belt conveyors

The collection of representative samples of grained materials is necessary to determine the quality parameters of this material. This operation is carried out in difficult conditions (dustiness, noise, atmospheric precipitation), which is a nuisance to the staff taking samples and adversely affects accuracy. In addition, the quality of representative samples is affected by the care of personnel. Therefore, to ensure a higher quality of sampling operations and to eliminate the work of people in difficult conditions, systems have been developed for the automatic collection of representative samples from conveyor belts. The work gives examples of automatic sampling systems equipped with samplers and other auxiliary devices. The work contains descriptions of construction, classifications, an example of selection and examples of construction solutions for automatic sampling of grained materials.