An Accident Waiting to Happen? The Shell Halex Reaction: How an Impurity Had a Catastrophic Effect on a Chemical Manufacturing Process

One of principles we strongly advocate during our process development and safety training courses is to establish the effect of impurities on a chemical reaction, particularly previously unknown impurities that might be introduced into an existing process stream via new batches of raw materials, reagents, or solvents. A change of supplier might mean a different way of making these precursors with different process impurities that could adversely affect your existing well optimised and robust synthetic chemistry.

On Tuesday 20^th March 1990 I was just starting out in my chemistry career- an undergraduate fortunate enough to have spent time in industry working on process development of potassium fluoride halex (halogen exchange) reactions used in the full-scale manufacture of fluoroaromatics.¹ This particular day had no significance for me. In the early hours of the morning (3:20am), 200 miles away in Stanlow, Ellesmere Port (North-West England)- an explosion ripped through the Shell fluoroaromatics plant injuring 6 of the 10 workers on site, one of whom died 3 weeks later. The plant was partially demolished, with debris and damage reported 500 metres away (Figure 1). The subsequent investigation provided some insight into the hidden dangers of unexpected impurities and how a normal day very rapidly turned into a nightmare. This article looks in detail at the chemistry being carried out on that day, the root causes of the accident and lessons we can take away.

The process running in the multi-purpose plant was batch conversion of 2,4-dichloronitrobenzene (DCNB) to 2,4-difluoronitrobenzene (DFNB) via a high temperature halogen exchange reaction using potassium fluoride in a polar aprotic solvent, in this case dimethylacetamide (DMAc) (Figure 2). A phase transfer catalyst, tetramethylammonium chloride (TMAC), was used to accelerate the somewhat sluggish reaction.² The inorganic by-product (potassium chloride) was removed from the crude reaction mixture by centrifugation and the remaining organic material purified by batch fractional distillation, separating the fluorinated product from unreacted starting material, by-products (including chlorofluoronitrobenzene), DMAc, and toluene used to wash the centrifuge. In a separate downstream reaction the DFNB product was hydrogenated to produce 2,4-difluoroaniline, an important raw material for used in the production of agrochemicals, pharmaceuticals and dyestuffs.³ The incident in Stanlow occurred during the halex step and didn’t involve the nitro group reduction.

15 tonne batches of DFNB were prepared in a 15m³ stirred reactor via addition of KF powder to a pre-mixed solution of TMAC in dry DMAc, then pumping in liquid DCNB. The molar ratios were 1 (DCNB): 2.3 (KF): 3 (DMAc): 0.14 (TMAC). The vessel contents were heated to reaction temperature, 145°C, over 3 hours and held at this temperature for an additional 14 hours. Pressurised water in the reactor jacket was intermittently applied to remove excess heat and maintain the reaction temperature.⁴ The pressure within the reactor was maintained at 0.2 barg with nitrogen. At no point during 14 years of plant operation had a pressure excursion been observed. As an engineering safety measure the vessel was fitted with a 5 barg relief valve venting to atmosphere. Vent sizing was calculated based on worst case scenario-cooling system failure.

Once the reaction was complete, the vessel contents were cooled, neutralised with NaHCO₃ and pumped into the centrifuge system where KCl (and residual KF) were separated and washed with toluene. A solution of the product in DMAc/toluene were pumped into holding tanks prior to batch-wise fractional distillation.

The development lab-based halex reactions I carried out early in my career at, what was at the time, ISC chemicals in Avonmouth, Bristol, UK (subsequently to become part of Rhodia/Rhone Poulenc) were very similar to those run at Stanlow. They typically were run at 100g scale, and differed from the Shell process in that sulfolane was used as solvent at a higher reaction temperature (200-250°C). The difluoro-product was distilled out of the vessel with sulfolane (which always solidified and blocked the condenser). The yields were generally high. A full mass balance based on fluoride/chloride analysis of the distillation residues and isolated organics confirmed that, not surprisingly, tar formation occurs- most likely via some kind of fluorodenitration process or thermal polymerisation.¹ These experiments (and I did a lot of them) were to look at the “activity”of KF obtained from different suppliers in a standard halex reaction (conversion of 2,4-DCNB to 2,4-DFNB using KF in sulfolane at >200°C), in this case without a phase transfer catalyst. The project was ostensibly to benchmark commercial KF against material we were planning to manufacture ourselves using HF produced on our site at Avonmouth. I found an old photograph of a 2,4-DFNB holding vessel in the decommissioned Avonmouth plant (Figure 3).⁵

Depending on the source of KF, I observed a big difference in reaction rate. This was initially attributed to variability in water content; however it was subsequently found to be a much deeper problem, with particle size and crystal morphology playing a major role. KF activity differences have been reported many times in the literature, the first detailed study was published by Langlois in 1996.¹ More recently a team at Merck, who were using KF in a halex process to make 2-fluoroadinene-an intermediate required in the manufacture of the reverse transcriptase inhibitor Islatrivir- used modern solid-state techniques to understand the effect in more detail.⁶ I spent a while working on the KF plant that we subsequently commissioned, and the spray-dried material we produced was as active as the best commercial material available at that time.

In Stanlow, the crude halex reaction product in the holding tanks was filtered again, and pumped, as required, into a batch distillation still with a fractionating column containing Sulzur Mellapak and fractionated at 200-225 mbar into 8 fractions (Figure 4). The fraction cuts were monitored by relative density changes and gas liquid chromatography (GLC).

The Stanlow fluoroaromatics plant was designed in 1975 originally to make chlorofluoronitro benzene. Surprisingly, DMSO was used as the solvent for halex reactions in the early years of operation. After 1981 this was replaced by DMAc for operational (and presumably safety) reasons.⁷ Before the DFNB manufacturing process was implemented, safety testing of the optimised halex reaction in this solvent system were carried out to confirm thermal stability at and above normal operating temperatures. Thermal stability of the filtered crude reaction mixture in the presence of iron powder and of DFNB distillation residues generated during purification was measured in an adiabatic (Sikarex) reaction calorimeter developed by Shell.⁸ Slow decomposition of the residues at 200°C evolved 48 kw/tonne of heat. Data from safety testing the reaction itself at 180°C was sufficient to provide confidence that a reaction temperature of 145°C would be safe. The process ran for many years without incident.

In the period 1982-1989 no difluoroaniline was manufactured in Stanlow, presumably due to a glut in market demand. During this period, Shell undertook lab-based re-optimisation of the process in attempt to increase throughput. The work was successful, a higher reaction temperature was identified (165°C v’s 145°C) which reduced the overall batch cycle time. As described above, the original process safety studies were carried out at 180°C so it was deemed unnecessary to repeat this work- the new reaction temperature being within the testing scope of the original assessment. All other reaction parameters remained unchanged. Subsequently 40 batches of DFNB were prepared at the new reaction temperature without incident.

Several weeks before the incident on 20^th March 1990, water being used to clean the centrifuge entered the crude DFNB distillation still. The water was removed by repeated azeotropic distillation with toluene. Although successful, water was found in a DMAc/DFNB intercut (Figure 4) and a previously unknown impurity (dimethylamino-fluoronitro benzene (and/or its regioisomer, Figure 6) was found in the DFNB stream. At this point I would have assumed a route-cause investigation would have been instigated addressing the origin of an impurity never seen before. The tainted DFNB was pumped into holding tanks for (subsequently unsuccessful) re-processing. The DMAc, however, was fully recovered and recycled. The solvent was stored, ready to be used in future halex reactions.

Two batches of DFNB, 41 and 42, were subsequently prepared before the recovered DMAc was introduced into the manufacturing process. The next batch, 43, contained mainly “old” DMAc with some solvent recovered from the re-processed “water incursion” re-work. This reaction proceeded uneventfully. With water incursion a distant memory, the next batch- number 44-was to seal the impending fate of the Stanlow halex plant.

Batch 44 used DMAc solely recycled from the water incursion incident. On 19^th March 1990 the reaction mixture began its heating cycle up to 165°C. But it didn’t stop there. The temperature continued to rise despite attempts to apply cooling through the reactor jacket. As the reaction began to thermally runaway the pressure began to rise (Figure 5). This was not apparent in the control room because no VDU pressure gauge was in use. An operator on the plant itself attempted to alert the team, however at that point the (inadequate) vessel venting system activated and the reactor exploded, with the harrowing consequences described at the beginning of the article.

So what happened? Not surprisingly an investigation was carried out using calorimetric and analytical tests, computer modelling and small-scale experiments in the development lab. Attempts to replicate the incident were meticulously conducted. Their findings and underlying chemistry is described nicely in a 1992 paper in J. Loss Prev. Process Ind. by van Reijendam et al.^9a

The gas release and pressure rise were something they’d not seen before and gave a clue as to the cause of the incident. Repeating the reaction under standard conditions showed noting that could explain the runaway. Spiking the reaction mixture with various additives that could have conceivably been in the process stream revealed that acetic acid promoted a rapid exothermic reaction with an adiabatic temperature rise to 240°C, at which temperature the nitroaromatics start to decompose. Analysis of the volatile components from this experiment revealed the presence of non-condensable gasses- ketene and carbon dioxide. The former is a highly flammable gas (b.p. -56°C) that can form explosive mixtures with air, making it especially hazardous in tanks or closed vessels. It cannot be transported or stored in gaseous form due to rapid dimerization and polymerisation. Not something you want to generate under thermal runaway conditions. Flames were seen coming out of the reactor before the explosion, consistent with spontaneous ketene ignition. The pressure in the vessel due to formation of these gases and the inability of the venting system to relieve the pressure resulted in catastrophic vessel failure. It’s been estimated that the energy released exceeded 25K MJ- more than could be generated by all the non-combustible chemistry. Based on the vessel contents of 4 tonnes of DCNB/DFNB and 6 tonnes of DMAc the decomposition energy should be around 8K MJ, the additional energy provided by ketene conflagration.

These observations still didn’t explain how the acetic was formed. In order to get to the heart of this we need to turn our attention to our old friend Le Chatelier. DMAc reacts with water to produce dimethylamine and acetic acid (Figure 6), with the equilibrium strongly in favour of DMAc. Heating DMAc/water mixtures to 160°C results in no detectable hydrolysis. If, however, an electrophilic scavenger is present in the system, DFNB- in our case, small equilibrium concentrations of dimethylamine react and shift the equilibrium to the right, magnifying the process and generating significant amounts of acetic acid (Figure 6).^9a,b

Once acetic acid is present, reaction with KF generates potassium acetate, reaction of which with DFNB produces chloronitrophenyl acetate (Figure 7), an intermediate unstable to the reaction conditions. Deprotonation of the acetyl group generates ketene, and/or acetyl fluoride, both of which can be hydrolysed to acetic acid. Potassium phenolate can dimerise giving the corresponding biaryl ether. The base strength of potassium fluoride is a central to this mechanism.¹⁰

DMAc and acetic acid form an azeotrope with a similar boiling point to pure DMAc, so solvent in the holding tanks, used in batch 44, would have contained undetected quantities of acetic acid. Ketene formation from phenyl acetate, either by thermolysis or formation of oxygen derived hydroxyl radicals, has been described in the literature.¹² Curiously ketene is a toxic component formed from esters during vaping.^12d

I found no mention of the fate of the phase transfer catalyst in any of the literature I read. Presumably, like the nitroaromatics, it too decomposed at high temperature. Interestingly, having read a bit on this topic, at high temperature in the presence of base, tetramethylammonium chloride can methylate phenols, with the generation of trimethylamine (b.p. 3°C) as a by-product (Figure 8).¹¹ This material, if it did form, would have contributed to the pressure increase, and, as with ketene, is an extremely flammable gas. It would be interesting to know if this was detected in ensuing subsequent stress-testing experiments.

There are several lessons to be learned from this incident. Re-cycling is a cornerstone of sustainable manufacturing; however we need to be proactively looking for build-up of impurities in these recycle streams. Analytical methods need to be fit for purpose and able to control the specification of input materials. Changes in technology and testing methods give additional insights into existing processes, and re-evaluation old processes in light of changing standards leads to increased process understanding- the ultimate goal of route development and optimisation.

Building on the seminal work of Gottlieb in the 1930’s,¹³ new methods for fluoro-halex chemistry continues to be developed, with several interesting papers having been published in this area. A few are listed in this reference.¹⁴

The halex reaction was the first industrial project I was involved at the start of my chemistry career, and I confess that at a young age process safety was not something I gave a great deal of thought to. The events described above re-enforce the dangers of complacency in running well established chemistry and the need for vigilance, especially if new batches of raw materials containing previously unknown impurities might have a critical impact on process safety.

This article is dedicated to the people who were impacted by the events in Stanlow 35 years ago.

See you next time.

References:

1. Fluorination of aromatic compounds by halogen exchange with fluoride anions (“halex reaction”): B. Langlois et al, Industrial chemical Library 1996, 8, 244-292; Fluoride ion as a nucleophile and leaving group in aromatic nucleophilic substitution reactions: V. Vlasov, Fluorine Chem. 1993, 61, 193-216
2. Development of PTC in halex reactions: a) Tetramethylammonium chloride as a selective and robust phase transfer catalyst in a solid-liquid halex reaction: the role of water: Y. Sasson et al, Chem. Commun. 1996, 297-298; b) Halex reactions of aromatic compounds catalysed by 2-azaallenium,carbophosphazenium, aminophosphonium and diphosphazenium salts: a comparative study: A. Pleschke et al, J. Fluorine Chem. 2004, 125, 1031-1038; b) Phosphonium ionic liquids as highly thermal stable and efficient phase transfer catalysts for solid–liquid halexreactions: c) S. Jaenicke et al, Catalysis Today 2012, 198, 300-304; d) Nucleophilic reactions using alkali metal fluorides activated by crown ethers and derivatives: S. Lee et al, Catalysts 2023, 13, 479
3. 2,4-DFNB Worldwide 2,4-difluoroaniline market research report 2025, forecast to 2031:  pmmarketresearch.com
4. An overview of the Shell fluoroaromatics plant explosion: D. Mooney, ICHEME Symposium Series 124, p381
5. 2,4-DFNB holding tank picture courtesy of whateversleft.com
6. Development of a commercial manufacturing route to 2-fluoroadenine, the key unnatural nucleobase in islatravir: C. Hong et al, Org. Process Res. Dev. 2021, 25, 395-494
7. Study on autocatalytic decomposition of dimethyl sulfoxide (DMSO): Y. Deguchi et al, Org. Process Res. Dev. 2020, 24, 1614-1620
8. Shell-modified Sikarex calorimeter as a screening tool for runaway reactions: J. Kars et al, Thermochimica Acta 1996, 289, 155-165
9. a) Explosion at the Shell fluoroaromatics plant at Stanlow March 1990- an investigation of the underlying chemistry: J. van Reijendam et al, J. Loss Prev. 1992, 5, 211-213; ibid Fluorine Chem. 1991, 1, p163; b) Shell Stanlow fluoroaromatics explosion- 20^th March 1990: assessment of the explosion and of the blast damage: A. Cates J. hazardous Materials 1992, 32, 1-39
10. Reactions of potassium fluoride in glacial acetic acid with chlorocarboxylic acids, amides, and chlorides. The effect of very strong hydrogen bonding on the nucleophilicity of the fluoride anion: J. Emsley et al, Chem. Soc., Dalton Trans., 1975, 2129-2134
11. a) Microwave-assisted methylation of phenols with tetramethylammonium chloride in the presence of K₂CO₃ or Cs₂CO₃: M. Kocevar et al, Tetrahedron 2008, 64, 11618-11624; b) Strategies for using quaternary ammonium salts as alternative reagents for alkylations: J. Templ et al, Chem. Eur. J. 2024, 30, e202400675
12. a) Vacuum thermolysis of aryl acetates. A mechanistic study: S. Nishida et al, Chem. Lett. 1974, 3, 1303-1304; b) High-temperature measurements of the reactions of OH with small methyl esters: methyl formate, methyl acetate, methyl propanoate and methyl butanoate: K. Y. Lam et al, Phys. Chem A 2012, 116, 12229-12241; c) The combustion of ethyl acetate, methyl propionate and iso-propyl acetate: D. Hoare et al, Comb. Flame 1970, 15, 61-70; d) Mechanistic rationale for ketene formation during dabbing and vaping: K. Munger et al, JACS Au 2024, 4, 2403-2410
13. The replacement of chlorine by fluorine in organic compounds: H. B. Gottlieb, J. Am. Chem. Soc., 1936, 58, 532
14. a) Developing efficient nucleophilic fluorination methods and application to substituted picolinate esters, M. Sanford et al, Org. Process Res. Dev. 2014, 18, 1045-1054; b) Anhydrous tetramethylammonium fluoride for room-temperature S_NAr fluorination: M. Sanford et al, J. Org. Chem. 2015, 80, 12137-12145; c) Multiple approaches to the in situ generation of anhydrous tetraalkylammonium fluoride salts for S_NAr fluorination reactions: ibid 2017, 82, 5020-5026; d) Acyl azolium fluorides for room temperature nucleophilic aromatic fluorination of chloro- and nitroarenes: M. Sanford et al, Org. Lett. 2015, 17, 1866-1869; e) Development of S_NAr nucleophilic fluorination: a fruitful academia-industry collaboration, Sanford et al Acc. Chem. Res. 2020, 53, 2372-2383; f) Nucleophilic routes to selectively fluorinated aromatics: D. Adams et al, Chem. Soc. Rev., 1999, 28, 225–231; g) Modern carbon–fluorine bond forming reactions for aryl fluoride synthesis, T. Ritter et al, Chem. Rev. 2015, 115, 612-633