Updated: November 7, 2026
Table of contents

A manufacturer wants to replace its established 316L stainless steel powder with a less expensive grade. The chemical composition meets the alloy specification, and the particle size distribution falls within the required range. However, the particles are less spherical, the oxygen content is higher, and spreading behavior differs from the current feedstock.
Whether the alternative powder can produce acceptable parts depends on its processing window. AI monitoring can help detect changes during manufacturing, but its signals cannot establish part quality without validation against measured defects. It cannot correct unsuitable powder characteristics.
Research into alternative and reused metal powders provides a useful starting point. The findings discussed below reflect specific materials, equipment, and experimental conditions, not universal acceptance limits.
Why Lower-Cost Powders Behave Differently
Metal powder prices reflect production methods, particle classification, material purity, and quality control. Vacuum induction inert gas atomization (VIGA) produces highly spherical, low-oxygen powders but carries a price premium. Conventional gas atomization and water atomization offer alternatives for alloys such as 316L stainless steel.
Fines increase cohesion and agglomeration, while coarse particles approaching or exceeding the effective recoater gap can jam beneath the blade and drag streaks through the layer. Packing density determines how much material occupies a given layer volume. Consequently, two powders with similar median particle sizes can produce quite different layers.
Chemical differences matter just as much. Water-atomized 316L generally contains more oxygen than VIGA and conventional gas-atomized grades, much of it associated with particle-surface oxides. These oxides can influence molten-metal wetting, affect melt-pool behavior, and contribute to oxide inclusions in the finished material. Research has demonstrated higher oxide-particle populations in LPBF specimens produced from water-atomized 316L [1, 2].
Powder chemistry can also change during manufacturing. Galicki and colleagues measured oxygen contents of 0.033 wt% in the original stainless steel feedstock and 0.100 wt% in ejected spatter. They also found increased oxygen in powder close to the processed regions of the bed [3].
A powder meeting nominal alloy and size specifications can still change chemistry in use, and behave differently during spreading and melting.
What the 316L Research Demonstrates
Riabov and colleagues compared 316L powder grades produced using VIGA, conventional gas atomization (GA), and water atomization (WA), including a modified water-atomized grade (WA-M) [1].
The researchers characterized particle morphology, density, and flow behavior before producing LPBF specimens at layer thicknesses from 20 to 80 µm.
At 20 µm, all four powders produced specimens with low residual porosity, ranging from 0.03% to 0.06%. The differences became pronounced as layer thickness increased.
| Layer thickness | VIGA | Conventional GA | WA | WA-M |
|---|---|---|---|---|
| 20 µm | 0.03% | 0.05% | 0.03% | 0.06% |
| 40 µm | 0.03% | 0.06% | 1.77% | 0.20% |
| 60 µm | 0.07% | 0.08% | 3.91% | 0.70% |
| 80 µm | 0.06% | 0.16% | 4.86% | 3.93% |
Mean residual porosity reported in Riabov et al., Appendix Table A3. The original study provides standard deviations and statistical comparisons. Processing parameters were changed with layer thickness.
The unmodified water-atomized powder reached 4.86% mean residual porosity at 80 µm, compared with just 0.16% for the conventional gas-atomized grade. Modifying the water-atomized powder improved performance at intermediate thicknesses, but did not eliminate the difference at 80 µm.
The mechanical tests tell a more detailed story. Specimens produced at 20 µm were evaluated for tensile properties and impact toughness. These tests covered VIGA, conventional GA, and unmodified WA powders. The modified water-atomized grade was not mechanically tested.
The conventional gas-atomized powder produced the highest static strength, while VIGA specimens showed the greatest elongation to fracture. Impact toughness was similar for VIGA and conventional GA but significantly lower for the unmodified water-atomized grade. Fatigue performance was not evaluated.
The water-atomized powder was therefore not inherently unprintable. Under the tested parameter combinations, however, its residual porosity increased substantially at higher layer thicknesses. Even where low porosity was achieved, the mechanical response was not identical.
Which Powder Measurements Predicted Printability?
The researchers also evaluated conventional Hall flow, powder rheometry, and rotating-drum measurements.
Bulk density, basic flow energy, and break energy differentiated the powder grades more effectively than Hall flow alone. These measurements showed stronger relationships with spreading performance and the resulting density of printed specimens.
A Hall funnel measures free discharge under gravity. A recoater forces particles to rearrange within a confined, moving layer. The forces and failure mechanisms differ.
Include bulk density and dynamic powder measurements when screening alternative feedstocks, then verify their relevance through printing trials.
For more detail on separating material effects from recoater problems, see Powder Spreading in Powder Bed Fusion: What Recoater Interaction Reveals About Layer Defects.
What In-Situ Monitoring Can Measure
LPBF monitoring systems use optical, thermal, and imaging sensors to record events during manufacturing.
Thermal cameras show how radiating regions develop and cool. Photodiodes record rapidly changing emissions, while coaxial systems observe radiation along the laser’s optical path.
These measurements provide information that commanded machine settings alone cannot reveal.
Thermal Measurements Require Careful Calibration
Dual-wavelength pyrometers estimate temperature from radiation collected in two spectral bands. They can reduce sensitivity to certain emissivity uncertainties, but their accuracy still depends on calibration and optical-system behavior.
NIST researchers demonstrated the importance of these limitations during commercial LPBF monitoring. Their high-speed thermographic system captured images at 1,800 frames per second [4].
The researchers also quantified how emissivity assumptions affect calculated temperatures. For a camera signal of 8,000 digital levels, assuming an emissivity of 0.5 when the true value was 0.2 produced an estimated temperature error of approximately 200 °C.
This was a calculated example of measurement uncertainty, not an observed process-temperature excursion.
Measurement position also matters.
Guo and colleagues investigated a coaxial dual-wavelength system in which galvanometer movement changed the angle of light entering narrowband filters. The resulting spectral shifts affected temperature measurement across the build plate. Their 2026 study developed an optical configuration to reduce this effect [5].
For powder comparisons, establish a sensor baseline before drawing conclusions from changes in signal intensity or calculated temperature. Otherwise, position-dependent instrument behavior can be mistaken for feedstock variability.
What Machine Learning Adds
Simple threshold-based monitoring detects when a signal crosses a predefined threshold. Machine learning can examine more complex combinations of signals, including changes in intensity, duration, sequence, and spatial distribution.
Anomaly detection identifies behavior that differs from a reference dataset. Defect prediction goes further by connecting signal patterns with measured pores, lack of fusion, or other defined defects.
Linking Melt-Pool Images to Actual Pores
Tao and colleagues used in-situ melt-pool images and X-ray computed tomography (X-CT) to study keyhole pore formation in Ti-6Al-4V [6].
The researchers deliberately induced keyhole conditions by changing laser power or scanning speed. They then compared monitoring data with pores identified through X-CT.
A significant difficulty was pore movement. A bubble could travel within the liquid melt pool before becoming trapped, so the final pore location did not always align directly with a single monitoring image.
The researchers addressed this by connecting multiple frames with each defect location. A Random Forest model using physics-informed features achieved an area under the receiver operating characteristic curve (AUROC) of 0.95 and an area under the precision-recall curve (AUPRC) of 0.92.
That performance belongs to one alloy, one defect mechanism, and deliberately induced keyholes. Transfer the approach to another powder or production process, and the detection relationship must be validated again.
What Unsupervised Learning Changes
Li and colleagues took a different approach in their 2026 study of thin-walled LPBF structures [7].
Their method used unsupervised learning to identify unusual patterns in melt-pool monitoring signals without requiring pore labels during model learning.
Accurately labeling pores in complex geometries is difficult and expensive. The model can learn characteristic signal patterns before those patterns are interpreted in relation to defects.
The authors reported an F1 score of 90.79% for identifying pore-prone regions. Although the learning approach was unsupervised, evaluating an F1 score requires comparing predictions with labeled reference classifications.
The reported result therefore concerns the classification of pore-prone regions against an evaluation reference. It is not a 90.79% probability of detecting every physical pore in a production component.
What Powder Reuse Reveals About Powder Testing
Powder reuse offers another example of how material properties can move in different directions.
Cavaliere and colleagues studied WE43 magnesium alloy through 18 successive LPBF printing cycles using a single 750 g powder batch without adding virgin material [8]. The powder was sieved after each cycle at 67 µm to remove oversized particles and agglomerates.
Repeated use preferentially depleted particles smaller than approximately 20 µm, while particle irregularity, satellite formation, and agglomeration increased.
The Hall flow measurement improved by more than 30%. In practical terms, the reused powder discharged faster through the funnel, reducing the measured flow time in seconds per 50 g.
The same material performed worse in a separate powder-spreadability test. At a 30 µm spread-layer thickness, the reused powder produced a significantly lower layer density than virgin powder. The larger proportion of irregular particles made thin-layer formation more difficult. At 60, 90, and 120 µm, the reduction in layer density was not statistically significant.
As with the 316L findings, the difference between powders became particularly apparent under specific layer conditions, although the two studies showed different responses to layer thickness.
Yet printed WE43 specimens retained average densities above 99%, and hardness remained largely unchanged. Faster funnel discharge did not mean better thin-layer spreading.
This creates a practical problem for monitoring. A model trained on virgin powder may flag changes in spreading behavior after reuse even when the resulting parts continue to meet their measured quality requirements. Without validation against actual defects, these alarms can increase inspection effort without improving acceptance decisions.
The implications for powder genealogy and controlled reuse are explored in Virgin-to-Reused Powder Blend Ratios in Aerospace PBF-LB Qualification.
How to Evaluate an Alternative Powder
A useful evaluation separates powder suitability from monitoring performance. The first determines whether the material produces acceptable parts. The second establishes whether the monitoring system provides reliable information about the process.
1. Establish the Reference Feedstock
Characterize the established powder and candidate material using a common sampling and testing procedure.
Record particle size distribution, morphology, oxygen and other relevant chemical constituents, apparent and tapped density, and appropriate flow properties. Include spreading tests where recoater behavior is a likely concern.
ISO/ASTM 52907:2019 provides a framework for metal powder characterization, sampling, and traceability [9].
2. Compare Build Performance
Begin with controlled builds using the same machine, parameters, geometry, build layout, and environmental conditions.
Compare layer quality and resulting defect populations. Repeat the comparison across independent builds and powder lots to expose variability that a single trial would miss.
If the candidate requires different processing parameters, establish and validate that operating window separately.
The 316L study illustrates why this matters: a powder producing 0.03% porosity at 20 µm reached 4.86% at 80 µm. The process settings and layer thickness cannot be separated from the feedstock decision.
3. Connect Monitoring Signals to Defects
Record monitoring data for the reference and candidate powders, then compare the indications with independently measured part defects.
Look for repeatable relationships between signals and defect type, location, and severity.
| Question | Required evidence |
|---|---|
| What does the system detect? | A defined anomaly or defect mechanism |
| Does it identify the correct location? | Spatially registered sensor and inspection data |
| How often does it miss a defect? | Detection sensitivity across independent builds |
| How often does it raise a false alarm? | False-positive rate under normal production conditions |
| Does changing the powder affect performance? | Validation across relevant feedstock lots and process conditions |
| What happens after an alarm? | Documented investigation and inspection criteria |
Avoid evaluating a model solely by splitting individual sensor records from the same build into training and test datasets. Independent builds and lots provide a more demanding assessment of performance.
4. Establish the Production Decision
Use the combined powder, process, and inspection evidence to determine whether the candidate remains within the accepted manufacturing envelope.
Where monitoring identifies a relevant anomaly, define whether it triggers additional inspection, a process review, or rejection.
Set these criteria before moving into routine production. A dashboard that reports thousands of events without specifying what operators should do with them creates information, not process control.
Where Qualification Requirements Apply
ISO/ASTM 52904:2024 addresses the operation and production control of metal powder bed fusion processes for critical applications, including requirements associated with feedstock, equipment, and process control [10].
More specific qualification requirements depend on the intended application.
For NASA spaceflight systems covered by NASA-STD-6030, Section 4.9 requires passive in-situ monitoring to be qualified before it can serve as a quantitative indicator of part quality for acceptance. Qualification must establish the physical basis of the measurement, its relationship with a defined defective process state, and detection reliability [11].
NASA explicitly encourages monitoring as a source of process-control data, even when it has not been qualified for part acceptance. Monitoring data can also guide targeted inspection. However, under the standard, even a qualified monitoring technique cannot completely replace nondestructive evaluation.
Section 1.4.2 also restricts adaptive closed-loop monitoring systems that alter the defined AM process without approved tailoring. This is a NASA qualification restriction, not a universal prohibition on adaptive LPBF control.
For other sectors, the manufacturer must establish the applicable customer, regulatory, and process requirements. The role assigned to monitoring should be supported by the corresponding validation evidence.
Does the Alternative Powder Reduce Production Cost?
A cheaper feedstock only improves economics when the cost of producing an accepted part decreases.
The comparison needs to include more than powder price:
- Feedstock purchase price, material consumption, and recovery
- Machine time, including any reduction in achievable layer thickness
- Powder testing, process development, and qualification
- Inspection and monitoring costs
- Scrap, rejected builds, and rework
An alternative powder that requires thinner layers may increase build time enough to eliminate its purchase-price advantage.
Monitoring can contribute financially by identifying problems earlier, reducing avoidable production losses, or making inspection more targeted. Those savings need to be demonstrated through actual production results.
The useful commercial metric is cost per accepted part, evaluated against the same technical requirements.
Conclusion
An alternative powder should enter production only after its processing window and resulting part properties have been validated. Monitoring can support that decision when its signals have a demonstrated relationship with relevant defects.
Feedstock characterization establishes what enters the machine. Process monitoring records what happens during manufacturing. Inspection and testing establish whether the outcome meets the required standard. A useful monitoring system connects these stages. It does not remove the need for them.
Frequently asked questions
References
- Riabov, D., Cordova, L., Hryha, E., and Bengtsson, S. (2022). Effect of powder variability on laser powder bed fusion processing and properties of 316L. European Journal of Materials, 2(1), 202–221.
- Riabov, D., Rashidi, M., Hryha, E., and Bengtsson, S. (2020). Effect of the powder feedstock on the oxide dispersion strengthening of 316L stainless steel produced by laser powder bed fusion. Materials Characterization, 169, 110582.
- Galicki, D., List, F., Babu, S. S., Plotkowski, A., Meyer, H. M., Seals, R., and Hayes, C. (2019). Localized Changes of Stainless Steel Powder Characteristics During Selective Laser Melting Additive Manufacturing. Metallurgical and Materials Transactions A, 50, 1582–1605.
- Lane, B., Moylan, S., Whitenton, E. P., and Ma, L. (2016). Thermographic measurements of the commercial laser powder bed fusion process at NIST. Rapid Prototyping Journal, 22(5), 778–787.
- Guo, F., et al. (2026). In-situ coaxial temperature monitoring of LPBF melt pool by a prism-filter dual-wavelength thermometry system. Additive Manufacturing, 122, 105194.
- Tao, Z., Thanki, A., Goossens, L., Witvrouw, A., Vrancken, B., and Dewulf, W. (2025). Data-driven keyhole pore detection in laser powder bed fusion: Integrating process insights with X-CT. Journal of Manufacturing Processes, 142, 293–316.
- Li, J., et al. (2026). Unsupervised learning-based identification of pore-prone regions via in-situ melt pool monitoring in thin-walled structures during laser powder bed fusion. Journal of Manufacturing Processes, 157, 1–14.
- Cavaliere, G. P., Shtender, V., Mellin, P., Persson, C., and D’Elia, F. (2025). Powder reuse in powder bed fusion-laser beam of WE43 magnesium alloy: towards sustainable manufacturing of biodegradable implants. Journal of Materials Research and Technology, 38, 5498–5510.
- ISO/ASTM 52907:2019. Additive manufacturing — Feedstock materials — Methods to characterize metal powders.
- ISO/ASTM 52904:2024. Additive manufacturing of metals — Process characteristics and performance — Metal powder bed fusion process to meet critical applications.
- NASA (2021). NASA-STD-6030: Additive Manufacturing Requirements for Spaceflight Systems. Sections 1.4.2 and 4.9.




