The research team, led by Trixi Meier of the University of Bremen and colleagues from TU Dortmund University and other German institutes, set out to answer a deceptively simple question: is cryogenic minimum quantity lubrication, or cMQL, always more sustainable than flood cooling? Their conclusion is emphatically no, at least not without careful optimization. The problem lies not in the machining itself, where the technology performs brilliantly, but in the hidden upstream burden of producing liquid CO2. Although carbon dioxide is a cheap by-product of ammonia production, compressing and cooling it into a liquid state adds 0.72 kilograms of CO2-equivalent emissions per kilogram of liquid CO2 delivered to the machine tool.

To understand why this matters, consider how much liquid CO2 a cMQL system consumes. In the milling experiments that formed the basis of the life cycle assessment, the system burned through 8 kilograms of liquid CO2 every hour, alongside 60 milliliters of mineral oil. Multiply that consumption rate by the liquefaction penalty and the numbers become staggering: the CO2 supply chain alone contributes roughly 5.8 kilograms of CO2-equivalent per hour of cutting, dwarfing the emissions of the machining process itself. Only about a third of the total emissions in the cMQL scenario actually come from manufacturing operations.

The technical case for cryogenic cooling, however, is genuinely impressive. In deep hole drilling experiments on Inconel 718, a nickel-based superalloy notorious for destroying cutting tools, the cMQL system achieved cutting lengths exceeding 10 meters, equivalent to more than 333 boreholes with a single 2-millimeter drill. Conventional flood cooling with a 6 percent emulsion managed only about 4.5 meters, or roughly 150 holes, while flood cooling with specialized deep hole drilling oil reached approximately 7.5 meters. The cryogenic approach kept drilling torque below 0.2 newton-meters and feed force under 400 newtons throughout, while the emulsion system pushed torque above 0.35 newton-meters and forces beyond 900 newtons as wear accelerated.

Milling of titanium alloy Ti6Al4V told a similar story. With cMQL, the tool survived at least 45 meters of cutting, compared with just 25 meters under flood cooling with 10 percent emulsion, and the researchers noted that even at 45 meters the cryogenically cooled tool had worn to less than a third of the flood-cooled tool’s terminal wear. Remarkably, when the cutting speed was pushed from 70 to 130 meters per minute, the cMQL tool still showed almost identical wear at the same cutting length, suggesting the technology could enable substantially higher productivity. Tests with different lubricant oils revealed that natural esters performed best, while some additives actually worsened wear, underscoring how sensitive the process is to fluid selection.

The life cycle assessment itself followed the ISO 14040 and 14044 framework, using one hour of cutting time as the functional unit and drawing on primary experimental data combined with the Ecoinvent 3.10 database. Under a narrow system boundary that ignored CO2 liquefaction, cMQL looked like a clear winner: 2.8 kilograms of CO2-equivalent per hour versus 5.2 for flood cooling, a 45 percent reduction, alongside lower water consumption of 0.62 versus 1.04 cubic meters and reduced human toxicity potential. Tool production dominated emissions in both scenarios, contributing 4.36 kilograms of CO2-equivalent in the flood-cooled case and 2.43 in the cryogenic case, because the longer tool life meant fewer tools were consumed per hour of machining.

But when the researchers expanded the boundary to include liquefaction, storage and transport of liquid CO2, the picture inverted dramatically. Total emissions for cMQL jumped to 8.3 kilograms of CO2-equivalent, a 59.6 percent increase over flood cooling. The researchers stress that this result is scenario-dependent rather than a universal verdict, yet the implications are hard to escape. Even in a hypothetical future where the entire liquefaction chain ran on renewable electricity, the improvement would be modest, because electricity accounts for only about 17.7 percent of the emissions of producing liquid CO2. Methane emissions and heat demand dominate the balance, meaning the specific emission factor would fall only from 0.72 to roughly 0.59 kilograms of CO2-equivalent per kilogram, leaving the cMQL system’s liquefaction burden at around 4.7 kilograms per hour instead of 5.8.

The authors argue that the literature on cryogenic machining has systematically avoided this reckoning. Many studies compare cryogenic methods only against dry machining or against each other, rather than against the flood cooling that remains the industrial standard, and few include any quantified environmental assessment at all. Some studies that did compare against flood cooling found no process performance advantage. The team therefore established three minimum criteria for a meaningful comparison: the industrial standard must serve as the reference, at least two different oils must be tested, and cutting parameters must be varied rather than simply copied from conventional practice. Applying these criteria to the published literature, they found that not a single existing study satisfied all three, which is why they conducted their own drilling and milling experiments.

So what would it take for cryogenic cooling to genuinely earn its green credentials? The researchers calculate that emissions must fall from 8.3 to at most 4.16 kilograms of CO2-equivalent per hour, roughly a 50 percent reduction, just to offer a 20 percent advantage over flood cooling that would justify the industrial risk of switching technologies. The most powerful lever is simply using less CO2. Current systems often run continuously even when the tool is not engaged, and fixed nozzle diameters deliver more coolant than the cutting process requires, causing wasteful overcooling and thermal shocks that can damage workpiece surfaces. Integrating the CO2 supply into the machine spindle, throttling the mass flow to match actual cutting energy demand, and synchronizing delivery with tool engagement through the CNC system could all dramatically cut consumption. Redesigned tools with modified flank faces and internal channels can also direct cooling precisely to the hottest wear zones, extending tool life further while using less medium.

There are additional dividends beyond the machine tool itself. Because CO2-based machining leaves components far less contaminated than emulsion-based flood cooling, downstream cleaning steps, which are energy- and chemical-intensive, could be reduced or eliminated entirely, though these benefits were not included in the current assessment. More ambitiously, the CO2 released during machining could be captured from a sealed, CO2-rich machine atmosphere and recycled through a centralized liquefaction plant, closing the material loop in facilities operating many cryogenic machines. The researchers are careful not to oversell their conclusions: sustainability claims for cutting fluid strategies, they write, cannot be generalized and must always rest on a comprehensive life cycle assessment with fully disclosed system boundaries. Their message to industry is ultimately pragmatic rather than dismissive. Liquid CO2 cooling is not a universal green solution, but in demanding applications such as drilling deep holes in superalloys or milling titanium, where it delivers dramatic tool life gains and cleaner parts, targeted optimization of the CO2 supply chain could yet transform a technology that currently emits more than it saves into one that genuinely delivers on its environmental promise.

Subject of Research: Life cycle assessment of cryogenic minimum quantity lubrication with liquid CO2 in machining

Article Title: Comparative life cycle assessment of cryogenic minimum quantity lubrication in machining including CO 2 liquefaction

Article References: Meier, T., Arafat, R., Blömeke, S., Jois, P. K., Saelzer, J., Sicking, M., Biermann, D., Hanenkamp, N., & Herrmann, C. (2026). Comparative life cycle assessment of cryogenic minimum quantity lubrication in machining including CO2 liquefaction. Cleaner Engineering and Technology, 35, Article 101329. https://doi.org/10.1016/j.clet.2026.101329

Image Credits: AI Generated

DOI: 10.1016/j.clet.2026.101329

Keywords: cryogenic machining, minimum quantity lubrication, life cycle assessment, CO2 liquefaction, flood cooling, cutting fluids, tool life, Inconel 718, Ti6Al4V, sustainable manufacturing, greenhouse gas emissions, machine tools

The study, published in Results in Engineering by Kazuki Kaneko and Zongze Li, addresses a fundamental blind spot in digital manufacturing. Computer-aided manufacturing (CAM) software can simulate geometry and generate tool paths from a CAD model, but it cannot simulate what happens inside the machine’s controller. The G-code is filtered by the CNC interpolator, feedforward controllers and feedback control loops before it ever reaches the servomotors, and these proprietary algorithms are closely guarded black boxes maintained by machine tool builders. Physical modeling of this chain is extraordinarily difficult precisely because the control algorithms are secret and the dynamics are deeply nonlinear. The result is that manufacturers often discover machining errors only after cutting — or resort to time-consuming air runs in which the machine executes the program without a workpiece just to observe its behavior.

The authors reframed the problem as one of sequence translation. From a data-processing perspective, G-code execution is a stream of discrete commands fed row by row into a black-box system that emits corresponding sequences of tool-path and cycle time errors. Predicting those errors for a given program is therefore fundamentally a sequence-to-sequence problem — exactly the kind of task at which Transformers excel. Unlike recurrent networks or long short-term memory (LSTM) models, which process sequences step by step, a Transformer uses multi-head self-attention to derive global relational weights across all positions simultaneously. This allows it to capture long-range temporal dependencies, such as how a sharp corner several blocks ahead causes the servo system to decelerate and how a large acceleration from a previous segment continues to influence subsequent commands as the transient response gradually decays.

The researchers drew a striking analogy between the Transformer’s internal machinery and the physics of a CNC controller. In their framing, the query represents the motion state of the current G-code command, while the keys and values carry error features of the surrounding tool path; scaled dot-product attention then assigns weights to nearby and distant blocks, mirroring the feedforward and feedback structure of a real servo loop. Position-wise feedforward networks introduce the nonlinear modeling capacity needed to emulate local servo behaviors such as PID control and friction, while residual connections — which pass inputs directly forward — resemble a closed-loop feedback system that focuses on predicting small dynamic deviations rather than reconstructing the entire tool path. Layer normalization, meanwhile, stabilizes the wildly different physical scales of position increments and feed rates, functioning much like a gain stabilizer in a servo drive.

Training such a model requires data that spans the machine’s full dynamic repertoire, so the team avoided reliance on specific workpiece geometries, which would limit generalization. Instead, they semi-randomly generated ten sets of training G-code combining local blending segments, where the path passes directly through control points, and global blending segments, where B-spline interpolation smooths the trajectory at constant feed rate. Displacements and feed rates were drawn randomly from defined ranges, producing tool paths with varied travel ranges and frequent direction changes. Each program was then executed on a real three-axis ROBODRILL machine equipped with a Fanuc Series 31i controller, and actual tool-path data were captured at a sampling frequency of 1 kHz using Fanuc’s SERVO Guide monitoring software. The training targets were thus experimentally measured servo responses rather than virtual simulations, grounding the digital twin in physical reality.

The model itself is an encoder-only Transformer with a model dimension of 512, eight layers and four attention heads, trained for 2000 epochs on a workstation with an RTX 4090 GPU. Input sequences consisted of nominal positional increments along the three feed axes plus the commanded feed rate, segmented into windows of 32 control points using a causal sliding window. The output comprised the three-dimensional tool-path error and the cumulative cycle time error at each point. Notably, the authors found that longer input sequences converged more slowly without improving final accuracy — a result that aligns neatly with servo physics, since the transient influence of any single command naturally fades over time. When compared against a Bi-LSTM model of equal dimensions trained on identical data, the Transformer converged in roughly 500 epochs, three times faster than its recurrent rival.

Validation was carried out on six G-code programs derived from two physical specimens: a 25-millimeter-radius hemispherical part machined along morphing spiral and bidirectional paths, and a complex artifact geometry, each run at nominal feed rates of 1000 and 3000 millimeters per minute. The results were decisive. Both models avoided large systematic bias, with mean prediction deviations generally within ±1 micrometer or millisecond, but the Transformer’s uncertainty bands were roughly 1.5 to 2 times tighter than the Bi-LSTM’s across all error components — and in some conditions nearly four times tighter. Its error distributions clustered markedly more sharply around zero, and while the Bi-LSTM’s cycle time root-mean-square error ballooned to about 8 milliseconds on the most demanding program against a mean error of 2.5 milliseconds, the Transformer suppressed such outliers, keeping RMSE close to MAE.

The comparison also revealed how differently the two architectures degrade under stress. Higher feed rates inherently reduce the accuracy of the controller’s feedforward and feedback loops, and the Bi-LSTM’s errors doubled or even tripled when the feed rate rose from 1000 to 3000 millimeters per minute. The Transformer, by contrast, showed far more graceful degradation, keeping mean errors below 2 micrometers or milliseconds even on the most challenging high-speed program. The authors attribute the residual errors that remain — small but nonzero — to highly random mechanical nonlinearities such as micro-friction variations and backlash during axis reversals, along with the quantization limits of encoder feedback at the 1 kHz sampling rate.

Perhaps the most industrially compelling result is speed. Predicting errors for an entire validation program took the Transformer less than 3.5 seconds, regardless of program complexity, whereas air-running the same programs on the machine took anywhere from roughly 38 to 202 seconds. Because prediction time depends only on G-code length, the trained model can instantly screen massive production programs offline, eliminating the need for preliminary air runs and sharpening both production scheduling and machining efficiency. The approach does not generalize across machines — each machine tool has unique controller and mechanical characteristics — but the authors emphasize that adapting to a new machine requires only executing the training G-codes, acquiring a fresh dataset and retraining, after which the model serves as a customized digital surrogate for that specific machine.

The work points toward a future in which every machine tool carries its own learned digital twin, capable of vetting programs before metal is ever cut. The team’s next steps include extending the model to account for cutting force-induced deflections of tool and workpiece, which superimpose on servo tracking errors in real machining; probing how specific G-code features such as control point density affect prediction performance; scaling the sequence-to-sequence methodology to five-axis simultaneous machining; and developing real-time servo signal acquisition for online error prediction during actual cutting. They also plan to build open-architecture experimental CNC systems, since proprietary commercial controllers block the integration of theoretical control equations needed for hybrid physical-digital modeling. If those ambitions pan out, the humble G-code file may one day arrive at the shop floor already annotated with a complete forecast of its own imperfections.

Subject of Research: Transformer-based deep learning prediction of tool-path and cycle time errors in CNC machining from G-code

Article Title: Prediction of tool-path and cycle time errors of CNC machining using a transformer-based deep learning model

Article References: Kaneko, K., & Li, Z. (2026). Prediction of tool-path and cycle time errors of CNC machining using a transformer-based deep learning model. Results in Engineering, 32, Article 112872. https://doi.org/10.1016/j.rineng.2026.112872

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.112872

Keywords: CNC machining, Transformer model, deep learning, digital twin, tool-path error, cycle time prediction, G-code, servo dynamics, Bi-LSTM comparison, precision manufacturing, sequence-to-sequence learning, machine tools